CA1158174A - Multiple variable threshold speech detector - Google Patents
Multiple variable threshold speech detectorInfo
- Publication number
- CA1158174A CA1158174A CA000368374A CA368374A CA1158174A CA 1158174 A CA1158174 A CA 1158174A CA 000368374 A CA000368374 A CA 000368374A CA 368374 A CA368374 A CA 368374A CA 1158174 A CA1158174 A CA 1158174A
- Authority
- CA
- Canada
- Prior art keywords
- signal
- level
- representation
- speech
- output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J3/00—Time-division multiplex systems
- H04J3/17—Time-division multiplex systems in which the transmission channel allotted to a first user may be taken away and re-allotted to a second user if the first user becomes inactive, e.g. TASI
- H04J3/175—Speech activity or inactivity detectors
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B3/00—Line transmission systems
- H04B3/02—Details
- H04B3/20—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
- Transmission Systems Not Characterized By The Medium Used For Transmission (AREA)
- Time-Division Multiplex Systems (AREA)
- Telephone Function (AREA)
Abstract
MAY, C. J. JR. 16 MULTIPLE VARIABLE THRESHOLD
SPEECH DETECTOR
Abstract of the Disclosure A speech detector uses a signal classifier to identify portions of a representation of the average magnitude of a group of signal samples indicative of either speech or noise. A controller in the signal classifier follows a four state sequence using appropriate time constants for signal measures in a variety of signal conditions in defining the speech and noise portions of the representation to provide adaptively variable decision levels. A speech definer compares the representation to a first decision level and the signal samples to a higher decision level to indicate the occurrence of speech signal activity when either decision level is exceeded. In a two way transmission arrangement, a receive trunk speech detector uses a stretcher to prevent adaption of the transmit speech detector thresholds when echo signals are present.
SPEECH DETECTOR
Abstract of the Disclosure A speech detector uses a signal classifier to identify portions of a representation of the average magnitude of a group of signal samples indicative of either speech or noise. A controller in the signal classifier follows a four state sequence using appropriate time constants for signal measures in a variety of signal conditions in defining the speech and noise portions of the representation to provide adaptively variable decision levels. A speech definer compares the representation to a first decision level and the signal samples to a higher decision level to indicate the occurrence of speech signal activity when either decision level is exceeded. In a two way transmission arrangement, a receive trunk speech detector uses a stretcher to prevent adaption of the transmit speech detector thresholds when echo signals are present.
Description
MAY, C. J. JR. 16 , ~
MULTIPLE VARIABLE` THRESHOLD
SPEECH DETECTOR
Background of t e Invention ; This invention relates to signal detecting arrangements and, more particularly, to arrangements for detecting speech activity in the presence of noise.
Speech detection arrangements are useful in a variety of communications systems in which speech transmission paths are established in response to the occurrence of speech signal activity. Some examples of the use of speech detection include speech interpolation and echo suppression break-in purposes. The signal measurement criteria or speech definition of conventional speech detectors is sufficiently deficient that it has been necessary to extend the speech indication heyond each signal satisfaction of the speech definition by a rather long hangover interval (100 milliseconds or more) to provide the perception of quality transmission over speech interpolation systems.
Speech detectors, particularly those used for speech interpolation, should ideally define the minimum time intervals during which a transmission path is needed by a customer, expressed in a percentage of total time called activity, so that the listener will perceive the connection as having a 'good" quality. Hence, a speech detector should be highly sensitive to the presence of Speech signals while at the same time remaining insensitive to non-speech signals. This may be achieved by an improved speech definition which enables the hangover interval to be minimized without degrading performance. Consequently, the speech definition of known arrangements for detecting speech has generally suffered from such limitations as undesirable speech clipping on the one hand and excessive activity on the other hand due to utilization of a poor speech definition coupled with excessive hangover and due '` `'; ,~,. ~
;:
MAY, C. J. JR. 16 ~, L ~
to noise sensitivity.
A primary object of this inventior. is to provide an improved speech definition which allows a substantial reduction in the duration of hangover without producing the aforementioned drawbacks.
Another object of the present invention is to provide an improved method and arrangement for detecting speech activity in the presence of noise wherein noise level estimates are independently derived while talker volume estimates are derived in relationship to the noise level estimate.
A related object is to provide a signal classification process using an average representation of several signal samples wherein the signal classification process assigns appropriate time constants to signal measures of the representation while identifying portions which constitute speech and which constitute noise.
The invention classifies a signal using its average representation to indicate when prescribed signal attributes are characteristic of either speech or noise. A
level estimator responsive to these indications provides a noise level estimate and then a first decision level. The first decision level is adapted by combining the noise level estimate and the results of a comparison between a signal measure of the speech portion of the representation and the current value of the first decision level. Each time the signal exceeds the first decision level, speech activity is indicated.
In some other aspects of the invention, speech activity is also indicated when the representation exceeds a second and lower decision level derived from the first decision level. The classifying means comprises measuring means for obtaining a peak and minimum values of the representation which a comparer utilizes to provide an output indicative of noise when the peak value is less than a prescribed constant times the minimum value. The level estimator compares the minimum to a stored long term . ~
;~ .
11~8~'7~
minimum and adjusts the latter by a prescribed amount to reduce inequality of the comparison~ The classifying means further comprises comparing means for providing an output indicative of speech when the peak value is greater than the sum of a predetermined factor times the minimum value plus a prescribed constant.
In some further aspects of the invention, the signal is sampled and digitally encoded so the level estimator includes a translator which uses the long term minimum to provide a noise level estimate compatible with the digitally encoded signal samples. The portion of the level estimator which provides the first decision level includes a comparer, an adjuster, a storer, and an adder.
The comparer produces an output indicative of the relative magnitude of the peak value of the representation and the first decision level. The adjuster uses this output to produce a positive increment when the peak value exceeds the first decision level and a negative increment when the first decision level is largër than the peak value. The storer accumulates and retains these increments and provides a level indicative of talker volume relative to - the noise level estimate. The adder combines the noise level estimate from the translator and the level indication to form the first decision level.
In still further aspects of the invention, echo protection is provided in a two-way arrangement. The arrangement prevents the utilization of echo signal energy in forming the adjustments of the variable threshold levels. Also, a stretcher or echo envelope generator provides an output whose amplitude corresponds to the greater of the representation of the signal energy in the received path or any other of previously occurred representations of the same type within a predetermined delay interval.
-~ 1 5 ~ 1 7 ~
- 3a -In accordance with an aspect of the invention there is provided apparatus for indicating the occurrence of speech in a signal indicative of both speech and noise, the apparatus including means for generating a represent-S ation of the average magnitude of the signal during amoving time interval; the apparatus being characterized by classifying means for receiving said representation and a noise level estimate, said classifying means generating a first output to indicate when said representation has prescribed attributes indicative of speech and as second output to indicate when said representation has prescribed attributes indicative of noise; level estimator means responsive to said first and second outputs and said representation, said level estimator means providing a lS noise level estimate using the portion of said represent-ation identified by the occurrence of said second output, said level estimator means providing a first decision level output by combining said noise level estimate and the portion of said representation defined by the occurrence of said first output in excess of a prescribed amount of said first decision level output; and comparing means for providing an output indicative of the occurrence of speech signal activity when said first decision level is exceeded by the signal.
Brief Description of the Drawing _ _ _ _ The above and other objectives and features of the invention may be fully appreciated from the following MAY, C. J. JR. 16 ~~ I 1 ~ ~ r7 D~
detailed description which should be considered in conjunction with the accompanying drawing.
FIG. 1 is a diagram of a basic speech detector constructed in accordance with the present invention.
FIG. 2 illustrates the internal circuitry of a signal classifier suitable for use in FIG. 1.
FIG. 3 depicts the circuitry of a level estimator suitable for use in E`IG. 1.
FIG. 4 is a state diagram which illustrates the operation of the controller used in FIG. 2.
FIG. 5 depicts the timing relationship between granularity pulses used in the operation of the controller of FIG. 2.
FIG. 6 illustrates an application of the invention wherein echo protection is utilized.
FIG. 7 is a circuit diagram of the type of stretcher employed in FIG. 6.
FIG. 8 illustrates the timing of pulse waveforms used in the operation of the circuit of FIG. 7.
Detailed Description In FIG. 1, transmit trunk 11 represents a typical teleeommunieations trunk whieh earries digital signals including speech. All of the signals on trunk 11 are presented for evaluation by the circuitry in FIG. 1 which forms an illustrative speech detector. The primary function of the speech detector of FIG. 1 is to signal at output 12 when speech signals are present on trunk 11.
Versions of the signal on trunk 11 produced via measure generator 13 are applied to variable threshold generator 14 and speech definer 16. Briefly, variable threshold generator 14 develops relatively long term, continuous, adjustable, levels which are used by speeeh definer 16 as deeision levels. When the deeision levels are exeeeded, an aetive speeeh indieation is produced at output 12 for the duration of the detection period plus an interval following the eessation of sueh speeeh activity previously referred to as hangover.
"
:.,, . :
, .:
.. .
MAY, C. J JR. 16 ~ 5 I'he hangover interval of the speech detector embodiment in FIG. 1 may comprise either a fixed interval of time or a variable interval of time or both. The improved speech definition of the present speech detector enables a relatively small fixed hangover of 16 ms to provide acceptable performance. The hangover interval also serves to bridge short gaps between speech intervals which, in turn, desirably reduces the number of service requests, i.e., the number of changes in state of the binary active speech identification signal at output 12.
In measure generator 13, the signal from transmit trunk 11 is filtered by high-pass filter 17 to suppress direct current offsets, power line crosstalk, and the like to ensure more accurate operation of the speech detector.
Measure generator 13 produces two outputs which are the single sample(SS) magnitude signal and the exponentially mapped past (EMP) signal which is an average representation of the signal power of energy. This average signal is produced by exponential weighting of past single sample magnitude signals in averager 18. The equivalent time constant of averager 18 is 16 ms and it may be viewed as the digital equivalent of an analog resistance capacitance (RC) low-pass filter. These two outputs from measure generator 13 provide the primary measures for the processing performed by the remaining circuitry in FIG. 1.
only the EMP signal is applied to variable threshold generator 14 which includes signal classifier 19 and level estimator 21. Signal classifier 19 may be expanded further to recognize additional signal types besides the speech and noise outputs for level estimator 21. It should be noted that the operation of signal classifier 19 and level estimator 21 may be considered to be bootstrapped by virtue of conductor 22 which feeds back a threshold signal (TN) from level estimator 21 to signal classifier 19. The two other outputs of level estimator 21 are the magnitude ~TM) decision level and the lower energy (TE) decision level~
MAY~ C J. JR. 16 '7 ~
whiCh are utilized by respective comparers 23 and 24 in speech definer 16. Por the other speech definer signal inputs, the single sample magnitude is applied to comparer 23 while the EMP signal is applied to comparer 24.
Comparer 23 serves to provide a rapid response to a loud talker while comparer 24 provides a slower, but sure response to a lower level talker. Reference 26 produces a fixed level for signal classifier 19 and comparer 27 of speech definer 16. The fixed level (FL) is used so that comparer 27 activates the speech detector for the typical -30 dBmO tone levels used for signaling purposes. This level corresponds in the ~ law coding format to level 32.
It should be pointed out that this feature is primarily a function of the system where the speech detector is used and, of course, may be modified or omitted without affecting the actual speech detection function. The outputs of comparers 23, 24 and 27 are logically combined in OR gate 28 to produce an output from the gate for speech definer 16 when the decision level of any comparer is exceeded.
After the output of gate 28 has ceased for a period of time, hangover timer 29 will have produced an output for the duration of the output of gate 28 extended by the hangover interval. Should the output of gate 28 reappear before the hangover interval lapses, hangover timer 29 will be reset and provide the bridging function which was previously mentioned. In any case, timer 29 will always extend the duration of the output of gate 28 by the prescribed duration of the hangover interval.
FIGS. 2 and 3 respectively depict the circuitry of signal classifier 19 and level estimator 21 of variable threshold generator 14 which develops dynamically adjustable decision levels for speech definer 16. Since this portion of the speech detector provides the basis of its operation and performance, generator 14 will be described in detail. Briefly, generator 14 is designed to take into account the particular talker and noise level on ~: :
MAY, C. J. JR. 16 the trunk currently being serviced to provide the best grade of service Eor all talkers within a relatively wide range of possible background noise levels and without producing excessive spurious activity indications at output 12. The specifics of the description to follow is based on the application dependent assumption that the digital encoding of the signals on trunk 11 is in accordance with the standard Unites States 8-bit ~ 255 format ( ~ law) wherein approximately 256 levels are used and the minimum negative and positive maximum levels of 127 are attained by the peaks of a +3 dBmO sinusoidal waveform.
However, those skilled in the art may readily apply the inventive principles to any desired encoding format and also to operate directly with analog signals.
In FIG. 2, the EMP average representation is applied to peak catcher and store 31 and minimum catcher and store 32 which serve respectively to extract and retain the peak (P) and minimum (M) of the EMP signal values.
Since the timing used to obtain these signal values is determinative on the outcome, controller 33 issues pulses for resetting stores 31 and 32 at appropriate times. For the time being, it is sufficient to describe controller 33 as being essentially a four state sequence machine which establishes appropriate time constants for each signal measure to match the variety of signaling conditions produced by a typical telecommunications trunk. These states are designated as IDLE, WAIT, NOISE ACTIVE and SPEECH/TONE ACTIVE. Each state has an associated conductor for indicating machine status to timing generator 34. The inhibit input to controller 33, which serves to suspend operation, will not be discussed further at this time since its function is used in the application depicted by FIG. 6.
Timing generator 34 serves as an interval timer which helps to control the state transitions that occur within controller 33. Additional information pertaining to controller 33 will be provided in the discussion of the state sequence diagram shown in FIG. 4. The EMP signal is MAY,C. J. JR. 16 also applied to comparer 36 which will produce a signal to indicate to controller 33 when the EMP signal is greater than a noise level of TN which is fed back from level estimator 21. The remaining signal input to controller 33 is provided by comparer 37. Each time the peak from store 31 exceeds the fixed level (FL) from reference 26 comparer 37 indicates same for controller 33. Comparers 38 and 39 each receive both the P and M signal values from stores 31 and 32. Comparer 38 provides a speech (SP) output to indicate a speech produced EMP signal when P>2M+16. Comparer 39, on the other hand, identifies a noise produced EMP signal with a NSE output when P<1.5M.
It is again noted that these particular ratios are a function of ~ law coding while the underlying principles are applicable to other coding formats.
FIG. 4 is a state diagram for describing the operation of controller 33 of FIG. 2. In other words, controller 33 is a particular one of what is known in the art as sequential machines which are usually described by way of state diagrams. See, for example, Donald R. Haring, Sequential-Circuit Synthesis: State Assignment Aspects (Cambridge, Mass.: The MIT Press 1966). As can be readily observed, there are the four previously identified states in FIG. 4. At the outset, it is to be understood that the operation of the circuitry of FIGS. 2 and 3 is associated with a particular one of the several trunks typically connected to a TASI System. Accordingly, these four states may be thought of as trunk status store states which each have individual time constants for associated signal measures that may be thought of as trunk signal history formed from accumulations of previous signal measures.
Further more, it is to be understood that the functioning of variable generator 14 is independent of the actual speech detection function provided by speech definer 16 of FIG. 1. However, generator 14 provides the variable dynamic threshold levels that the latter uses in speech detection.
. ,." .i ;:
,, ~ ` ' .
MAY, C. J. JR. 16 Ll~17~
Initially, controller 33 starts in the IDLE state where a counter within timing generator 34 (timer 1, T01) provides pulses to define intervals of 256 ms wherein the peak, P, and the minimum, M, of the E~P of signal energy presumed to be indicative of noise are directly measured and stored to characterized this noise. Also, the value of L, the long term minimum of M, may be incremented one unit in 1/4 second intervals in the direction of L = M should this be necessary. Another function which occurs in the IDLE state and between connections of the trunk is that K, the talker level estimate, is incremented one unit every 16 seconds to make it equal to a typical talker within a rather large range (~45dB.) defined by upper and lower limits. Each unit change is equivalent to a 3 dB level shift. In particular, high thresholds left over from loud talkers are reduced while low thresholds left over from weak talkers are increased in anticipation of a new talker at a more likely talker level. This operation of characterizing noise only takes place until the EMP level exceeds the TN level when controller 33 switches to the WAIT
state.
In the WAIT state, the first event which may possibly occur is that P of the EMP energy average exceeds a high fixed threshold of ~ law level 32 which corresponds to about -30 dBmO. Since this high energy level is assumed not to be noise, controller 33 will switch to the TALKER
ACTIVE state. If this does not occur, the noise from the trunk which produced the state transition most likely was a transient condition, such as a pop or a click, then timer l(T01) will time out in 64 ms and switch controller 33 to the NOISE ACTIVE state. However, the values of P and M obtained from this signal condition that produced this transition are stored and used to adjust L.
While in NOISE ACTIVE, another timer in timing generator 34 (timer 2) is triggered by signal transitions to produce a pulse repetition rate referred to as granularity, (GX) which is divided into two slower pulse rates designated as .:
MAY, C. J. JR. 16 1 ~ 5~
granularity even, (GE), and granu:Larity odd, (GO~. As illustrated in FIG. 5, the first pulse to occur of the two slower pulse rates after the signal transition is the GE
pulse. The letter S designates the spacing or timing interval of the pulse repetition rate. It will become evident that a variety of pulse repetition rates are utilized to provide different time constants appropriate to the signal measure being performed during the several signaling conditions. In each case, the designated timing serves to identify the appropriate pulse repetition rate.
Now if the noise signal previously referred to as a click or pop was produced by the trunk being switched for a new call with a higher background noise level, the previously experienced M and P will be reset at the occurrence of GE and GO pulses respectively in 512 ms intervals while controller 33 is in the NOISE ACTIVE state.
Also the GC pulses will serve for updating L in 1/~ second intervals. After the levels of M and P are used to adjust L and 256 ms elapse without the signal energy exceeding the TN noise threshold, controller 33 will switch back to the initial IDLE state. The 256 ms interval is provided by timer 1 to provide a hangover or bridging function. While in the NOISE ACTIVE state the ratio of P/M is calculated to ascertain the occurrence of a low level speech signal. If the ratio of P/M is high enough to indicate speech, i.e., P>2M + 16 or the value of P exceeds the fixed level here taken as ~ law level 32, controller 33 will switch to the TALKER ACTIVE state. The state path including IDLE, WAIT, NOISE ACTIVE and returning to IDLE, however, may typically occur a number of times without entering the TALKER ACTIVE
state due to various types of activities within the telephone plant. However, after following this path, the variable thresholds of Tl~ and TE are better reference levels for use in detecting speech activity.
When controller 33 switches to the TALKER ACTIVE
state from either the WAIT or NOISE ACTIVE states, timer 2 of timing generator 34 provides longer duration timing .
, .
MAY, C. J. JR. 16 intervals for the measures associated with speech activity.
These include resetting M and P upon the occurrence of respective granularity pulses GE and GO at 2 second intervals. This state is given a certain latitude of stability by timer 1, which provides 256 ms of hangover by resetting timer 1 any time the signal energy exceeds the TN
noise threshold during the hangover interval. Also, K is updated at one second intervals defined by the GX pulses and L is updated at two second intervals defined by the GE
pulse train. If the signal happens to be a tone, the P/M
ratio will be lower than 1.5 and no K update occurs. In addition, if the level of the pack, P, is less than the fixed threshold, and the signal is classified as not speech, a state transition results to the NOISE ACTIVE
state upon the occurrence of -the next GX pulse. If the signal level drops below the noise threshold levels for a period of 256 ms, controller 33 will switch back to the IDLE state and reset the values of P and M obtained from the current noise levels.
In FIG. 3, level estimator 21 receives the outputs from FIG. 2 designated P, M, SP, and NSE. In addition, there are outputs directly from controller 33 which are used for updating. These outputs are T update noise or TUL and T update speech or TUK. In level estimator 21, comparer 41 receives the M signal and the noise level L signal which generally corresponds to the long term minimum of EMP. The value of L is obtained from noise level store or NL store 43 which is incremented by noise control 42 when it is clocked at 1/4 second intervals via the TUL line from controller 33. From the comparison of M and L at the end of each update interval the value of L in store 43 may not be adjusted at all or adjusted one unit in either a positive or negative manner in the direction to obtain L = M. l`he other input to noise control serves as an enabling signal. l'his is the noise (NSE) signal obtained frorn comparer 39 of FIG. 2 which signals the presence of a noise signal when there is a low MAY, C J. JR. 16 ratio in the comparison of the current peak (P) to minimum (M) signal values. Each adjustment incrernent is equivalent to approximately a 1 dB change in the value of L retained by store 43. I`his stored value is then altered by translator 44 to form the noise threshold or TN signal value before application to adder 46. This alteration or adjustment is primarily a function of the coding characteristic used to form the digital signal in trunk 11.
For ~ law, translator 44 multiplies L by 1.5 and then adds one to form T~.
For the speech signal levels comparer 47, talk volume control 48, K store 49, and translator 51 perform a very similar operation. Since in this case the output of adder 46, designated TM, is the sum of the outputs from translators 44 and 51 and provides one input to comparer 47, the comparison must be again related to the properties of the coding format used in providing the digital signal. For a completely linear coding format, TM
would be compared with P times a factor slightly greater than one. For a strictly logarithmic coding format, TM
would be compared with P minus a constant dB value. In the case of ~ law, a piecewise comparison is made since low levels are linearly related while high levels are linear on a logarithmic scale. When 2P is less than forty-eight, the value of K is eventually decremented to zero at which point TM corresponds solely to the noise level. In the range where 2P is greater than forty-eight but less then eighty, T~ is compared with 2P-16. Then the value of K is incremented when TM is smaller and decremented when TN is compared to 2P-16. When P is larger than eighty, TM is compared to 2P-24 and the value of K will be increased when TM is smaller and decreased when T~l is larger than 2P-24.
The output of comparer 47 may, as previously stated, signal no adjustment as a result of these comparisons.
Another input to control 48 is the SP signal of comparer 38 which enables the adjustment only when the P/M
ratio is indicative of speech. Each adjustment increment MAY, C. J JR. 16 'Il~l'i'~
to the value in K store 49, is equivalent to about a 3 dB
change in value at the output of translator 51 which provides a level, designated as TS/N, since it is a ratio of the talker level to the noise level. For the ~ law coding characteristic, translator 51 actually performs a multiplication by a factor of eight.
Accordingly, the latter circuitry forms an estimate of talker level derived from the peak of the average representation (P) normalized by L the noise level estimate which is combined in adder 46 whose output is 'r~.
T~ is compared by comparer 23 of FIG. 1 directly with the single sample magnitude. T~ is also applied to translator 52 whose output is TE a lower decision level for comparison to the EMP (signal power or energy) by comparer 24. This latter comparison insures a sure response to lower level talking signals while the former comparison provides a fast response to higher level talking signals.
For ~ law, translator 52 divides by a factor of four.
FIG. 6 depicts the application of the speech detector embodiment of FIG. 1 in a typical two-way transmission environment subject to echo in the form of signal leakage on transmit trunk 111 from receive trunk 115. The basic speech detector for transmit trunk 111 comprises measure generator 113, variable threshold generator 114, transmit speech definer 116 and reference 126 which function to provide an activity signal on conductor 112 indicative of the occurrence of speech for echo protector 131. Incidentally, the reference numerals in FIG. 6 whose last two digits correspond to FIG. 1 are used to designate identical components.
It should be pointed out that in system applications, such as Time Assignment Speech Interpolation (TASI) or Digital Speech Interpolation (DSI), the value of L (noise level estimate) and the value of ~M (talker level estimate) are useful measures. For example, the value of I
can be used to provide noise fill to prevent the perception of disconnects in a TASI arrangement. The remaining . ' MAY, C. J. JR. 16 components in E`IG. 6, namely, measure generator 132, stretcher 133, receive speech definer 134, break-in definer 136 along with echo protector 131 serve as a receive speech detector for interrupting the operation of the transmit speech detector. This enables the correct designation of intervals for accurate decision level determination. In other words, adaption of decision level should only occur using transmit signals not contaminated by echo leakage signals. Furthermore, the ability to detect expected echo energy prevents a request for trunk service if only echo is present and also allows the transmit speech detector to request trunk service in spite of the presence of echo. It should also be pointed out that the arrangement of FIG. 6 is required even though some form of echo protection is used in the transmission environment since echo protection techniques typically will not always eliminate echoes.
Aside from the basic operation of the transmit speech detector which remains primarily the same as before, the differences in operation attributed to the components used in the receive portion of the arrangement of FIG. 6 will be described. Measure generator 132 provides a receive exponentially mapped past (RRMP~ signal derived from receive trunk 115. Stretcher 133 uses this signal at its input and a version of it internally delayed by about 24 ms to provide a stretched REMP(SREMP) signal output whose amplitude at any instant corresponds to the greater of current REMP signal and any other previous REMP signal in the designated delay interval. Another and, perhaps, more descriptive term for stretcher 133 is an echo envelope generator. The delay of stretcher 133 takes into account the signal propagation experienced as the receive signal travels over the leakage path (indicated by the dashed arrow in FIG. 6) of a remotely located hybrid transfor-ner where the signals of trunks 111 and 115 are combined for two wire transmission. This is the signal path which must be traveled by the receive signal before it appears on MAY, C. J. JR. 16 transmit trunk 111. Break-in definer 136 compares the TEMP
and SREMP signal levels obtained from respective measure generator 113 and stretcher 133. Definer 136 issues a signal to indicate that the transmit energy exceeds the stretched and delayed receive energy. The break-in signal is supplied to echo protector 133 along with the output of speech definer 116.
I'he output of measure generator 132 is also applied to speech definer 134 whose output indicates when the input exceeds a fixed threshold representative of minimal but significant signal energy. Thus the speech detector arrangement for the receive signal need only be a simple energy detector which serves to indicate the presence of significant signal energy whether it constitutes speech and/or tone. This signal is applied to the inhibit input of generator 114 to stop the adaptation of the decision levels when the signal levels of trunk 111 are contaminated by echo signal energy.
Echo protector 131 provides an output designated as trunk needs service (TNS) to indicate the occurrence of active speech signals on trunk 111. This output is produced when the signal on conductor 112 indicates speech signal activity and definer 136 produces a break-in (BI) signal. If the activity signal ceases, or BI ceases then the TNS signal switches to a level indicative of an idle condition. When the TEMP>SREMP ceases, the BI signal will switch to the idle state only after the termination of a hangover interval or speech activity ceases. A hangover duration on the order of 256 ms has proven to provide satisfactory performance.
FIG. 7 depicts the circuit diagram of stretcher 133 in FIG. 6. The vertical dashed lines in FIG. 7 partition the circuit into an input section which is a peak catcher, a middle section which saves the last A peaks, and an output section which selects the greater peak from the other two sections. To input 141, delay, ~ , 142 and comparer, C, 143 are connected for application of the REMP
MAY, C. J. JR. 16 11~81 ~
Signal. Delay 142 has an enable terminal connected for activation by the output of OR gate 144. Gate 144 forces delay 142 to accept the REMP signal each time a Gl pulse occurs. FIG. 8 illustrates the GT pulse and its relationship to the other pulses used to operate the circuit of FIG. 7. rrhe output of delay 142 is connected to the other input of comparer 143. Comparer 143 provides an output signal to enable delay 142 via gate 144 when the level of input REMP signal exceeds the signal level stored in delay 142. This condition is indicated by the expression I~>NEW assigned to the output of comparer 143 in FIG. 7.
The function of the input section is to follow the rising peaks or slopes of REMP signal and then to retain the maximum peak value for the duration of the ~ interval. The increasing signal level is provided access to delay 142 by virtue of the output of comparer 143. In particular, comparer 143 produces this output signal when the signal level at input 141 exceeds the NEW level provided by the present signal value stored in delay 142. When this condition terminates, delay 142 holds the peak until the next G ~ pulse occurs.
l'he middle section of the circuit of FIG. 7 includes delay line 146 which has A-l units of delay and a signal comparison portion which includes the circuitry between selector 147 and selector 148. In this circuitry, two signal paths are provided. One signal path extends from the output of selector 147 via delay 149 to a first signal input of selector 148. The other signal path is line 151 extending from the output of selector 147 directly to the other signal input of selector 148. Selector 147 has a control input to which is applied the GT pulse. In selector 147, the application of the Gl pulse completes the signal path from the output of delay 142 to the input of delay 149. Similarlyr the occurrence of the high level signal, TA, at tne control input of selector 148 via OR
gate 153 completes a conductive path from the output of ;
MAY~ C. J. JR. 16 delay l4g to the input of delay line 146. This transfers the contents QA of delay 149 to the Ql unit of delay line 146 but not to delay 152 since Gl is absent. When the control signal is low, the other signal input of selectors 147 and 148 is coupled to their respective outputs.
As can be observed from E`IG. 7, the control input of selector 148 is connected to the output of OR gate 153 and OR gate 154 is associated with the enable input of delay 149. Accordingly, the occurrence of ~A pulse not only enables delay 149 to accept its signal input but also makes its output available at the output of selector 148 for delay line 146. The other input of gate 153 is connected to comparer 156 which enables the gate when the output level of delay 149 is greater than the signal level on line 151.
The purpose of the middle section of the circuit of FIG. 7 is to store the A past I interval peaks. During the G~ pulse the input to delay 149 is the NEW signal from the output of delay 142 while the output of selector 148 is the larger of the value retained in delay 149 or the NEW
signal on line 151 due to the operation of comparer 156.
During the portion of the pulse Tl-Ta not coextensive with the TA pulse, the input to delay 149 is QOLD, the previously saved I interval peaks stored in delay 146. Now the output of delay 149 is the previously stored NEW value since the output of gate 154 is not enabling the input to the delay at this time. Accordingly, the output of selector 148 is the larger of the old stored values or NEW
which is also due to the operation of comparer 156. When the TA pulse occurs, the contents of delay 149 is applied to the first unit of delay line 146 while delay 149 is being loaded with the contents of the last unit in delay line 146 which is the oldest of the previous stored values.
This process will be repeated upon occurrence of the next Gl pulse.
MAY, C. J. JR. 16 The output section of the circuitry of FIG. 7 includes previously mentioned delay 152 which provides a signal input to comparer 157 and selector 158.
Attenuator 159 is connected to the output of selector 158 to provide the SREMP signal with a loss adequate to compensate for the minimum expected return loss of the echo path symbolically indicated in FIG. 6. For TASI systems, this loss is about 6 db. The other signal inputs to comparer 157 and selector 158 are provided by the output of delay 142 which is the NEW signal. Comparer 157 provides the control input for selector 158 so that the selector's output always corresponds to the larger of the NE~ Signal or the OLD signal.
For a typical application in a TASI system specific values will provide satisfactory performance. For example, the precision may be defined in terms of time accuracy, l=4 ms, and bit accuracy, N=8 bits. A total delay of 24 ms is provided with A=4 units in delay line 146 along with the presence of delays 142 and 152 in the input and output sections of stretcher 133. Of course, in other applications different specific values may be readily selected by those working in the art in the art to provide suitable performance.
What has been described hereinabove is but illustrative of the principles of my invention. For example, it will be appreciated that applicant~s speech detector, though shown and described in connection with a single transmission channel for purposes of describing the invention, may be shared advantageously among a plurality of transmission channels on a time division multiplexed basis, each channel being connected to the speech detector during a distinct time slot. Numerous and varied other arrangements may be derived by those skilled in the art without departing from the spirit and scope of the invention.
MULTIPLE VARIABLE` THRESHOLD
SPEECH DETECTOR
Background of t e Invention ; This invention relates to signal detecting arrangements and, more particularly, to arrangements for detecting speech activity in the presence of noise.
Speech detection arrangements are useful in a variety of communications systems in which speech transmission paths are established in response to the occurrence of speech signal activity. Some examples of the use of speech detection include speech interpolation and echo suppression break-in purposes. The signal measurement criteria or speech definition of conventional speech detectors is sufficiently deficient that it has been necessary to extend the speech indication heyond each signal satisfaction of the speech definition by a rather long hangover interval (100 milliseconds or more) to provide the perception of quality transmission over speech interpolation systems.
Speech detectors, particularly those used for speech interpolation, should ideally define the minimum time intervals during which a transmission path is needed by a customer, expressed in a percentage of total time called activity, so that the listener will perceive the connection as having a 'good" quality. Hence, a speech detector should be highly sensitive to the presence of Speech signals while at the same time remaining insensitive to non-speech signals. This may be achieved by an improved speech definition which enables the hangover interval to be minimized without degrading performance. Consequently, the speech definition of known arrangements for detecting speech has generally suffered from such limitations as undesirable speech clipping on the one hand and excessive activity on the other hand due to utilization of a poor speech definition coupled with excessive hangover and due '` `'; ,~,. ~
;:
MAY, C. J. JR. 16 ~, L ~
to noise sensitivity.
A primary object of this inventior. is to provide an improved speech definition which allows a substantial reduction in the duration of hangover without producing the aforementioned drawbacks.
Another object of the present invention is to provide an improved method and arrangement for detecting speech activity in the presence of noise wherein noise level estimates are independently derived while talker volume estimates are derived in relationship to the noise level estimate.
A related object is to provide a signal classification process using an average representation of several signal samples wherein the signal classification process assigns appropriate time constants to signal measures of the representation while identifying portions which constitute speech and which constitute noise.
The invention classifies a signal using its average representation to indicate when prescribed signal attributes are characteristic of either speech or noise. A
level estimator responsive to these indications provides a noise level estimate and then a first decision level. The first decision level is adapted by combining the noise level estimate and the results of a comparison between a signal measure of the speech portion of the representation and the current value of the first decision level. Each time the signal exceeds the first decision level, speech activity is indicated.
In some other aspects of the invention, speech activity is also indicated when the representation exceeds a second and lower decision level derived from the first decision level. The classifying means comprises measuring means for obtaining a peak and minimum values of the representation which a comparer utilizes to provide an output indicative of noise when the peak value is less than a prescribed constant times the minimum value. The level estimator compares the minimum to a stored long term . ~
;~ .
11~8~'7~
minimum and adjusts the latter by a prescribed amount to reduce inequality of the comparison~ The classifying means further comprises comparing means for providing an output indicative of speech when the peak value is greater than the sum of a predetermined factor times the minimum value plus a prescribed constant.
In some further aspects of the invention, the signal is sampled and digitally encoded so the level estimator includes a translator which uses the long term minimum to provide a noise level estimate compatible with the digitally encoded signal samples. The portion of the level estimator which provides the first decision level includes a comparer, an adjuster, a storer, and an adder.
The comparer produces an output indicative of the relative magnitude of the peak value of the representation and the first decision level. The adjuster uses this output to produce a positive increment when the peak value exceeds the first decision level and a negative increment when the first decision level is largër than the peak value. The storer accumulates and retains these increments and provides a level indicative of talker volume relative to - the noise level estimate. The adder combines the noise level estimate from the translator and the level indication to form the first decision level.
In still further aspects of the invention, echo protection is provided in a two-way arrangement. The arrangement prevents the utilization of echo signal energy in forming the adjustments of the variable threshold levels. Also, a stretcher or echo envelope generator provides an output whose amplitude corresponds to the greater of the representation of the signal energy in the received path or any other of previously occurred representations of the same type within a predetermined delay interval.
-~ 1 5 ~ 1 7 ~
- 3a -In accordance with an aspect of the invention there is provided apparatus for indicating the occurrence of speech in a signal indicative of both speech and noise, the apparatus including means for generating a represent-S ation of the average magnitude of the signal during amoving time interval; the apparatus being characterized by classifying means for receiving said representation and a noise level estimate, said classifying means generating a first output to indicate when said representation has prescribed attributes indicative of speech and as second output to indicate when said representation has prescribed attributes indicative of noise; level estimator means responsive to said first and second outputs and said representation, said level estimator means providing a lS noise level estimate using the portion of said represent-ation identified by the occurrence of said second output, said level estimator means providing a first decision level output by combining said noise level estimate and the portion of said representation defined by the occurrence of said first output in excess of a prescribed amount of said first decision level output; and comparing means for providing an output indicative of the occurrence of speech signal activity when said first decision level is exceeded by the signal.
Brief Description of the Drawing _ _ _ _ The above and other objectives and features of the invention may be fully appreciated from the following MAY, C. J. JR. 16 ~~ I 1 ~ ~ r7 D~
detailed description which should be considered in conjunction with the accompanying drawing.
FIG. 1 is a diagram of a basic speech detector constructed in accordance with the present invention.
FIG. 2 illustrates the internal circuitry of a signal classifier suitable for use in FIG. 1.
FIG. 3 depicts the circuitry of a level estimator suitable for use in E`IG. 1.
FIG. 4 is a state diagram which illustrates the operation of the controller used in FIG. 2.
FIG. 5 depicts the timing relationship between granularity pulses used in the operation of the controller of FIG. 2.
FIG. 6 illustrates an application of the invention wherein echo protection is utilized.
FIG. 7 is a circuit diagram of the type of stretcher employed in FIG. 6.
FIG. 8 illustrates the timing of pulse waveforms used in the operation of the circuit of FIG. 7.
Detailed Description In FIG. 1, transmit trunk 11 represents a typical teleeommunieations trunk whieh earries digital signals including speech. All of the signals on trunk 11 are presented for evaluation by the circuitry in FIG. 1 which forms an illustrative speech detector. The primary function of the speech detector of FIG. 1 is to signal at output 12 when speech signals are present on trunk 11.
Versions of the signal on trunk 11 produced via measure generator 13 are applied to variable threshold generator 14 and speech definer 16. Briefly, variable threshold generator 14 develops relatively long term, continuous, adjustable, levels which are used by speeeh definer 16 as deeision levels. When the deeision levels are exeeeded, an aetive speeeh indieation is produced at output 12 for the duration of the detection period plus an interval following the eessation of sueh speeeh activity previously referred to as hangover.
"
:.,, . :
, .:
.. .
MAY, C. J JR. 16 ~ 5 I'he hangover interval of the speech detector embodiment in FIG. 1 may comprise either a fixed interval of time or a variable interval of time or both. The improved speech definition of the present speech detector enables a relatively small fixed hangover of 16 ms to provide acceptable performance. The hangover interval also serves to bridge short gaps between speech intervals which, in turn, desirably reduces the number of service requests, i.e., the number of changes in state of the binary active speech identification signal at output 12.
In measure generator 13, the signal from transmit trunk 11 is filtered by high-pass filter 17 to suppress direct current offsets, power line crosstalk, and the like to ensure more accurate operation of the speech detector.
Measure generator 13 produces two outputs which are the single sample(SS) magnitude signal and the exponentially mapped past (EMP) signal which is an average representation of the signal power of energy. This average signal is produced by exponential weighting of past single sample magnitude signals in averager 18. The equivalent time constant of averager 18 is 16 ms and it may be viewed as the digital equivalent of an analog resistance capacitance (RC) low-pass filter. These two outputs from measure generator 13 provide the primary measures for the processing performed by the remaining circuitry in FIG. 1.
only the EMP signal is applied to variable threshold generator 14 which includes signal classifier 19 and level estimator 21. Signal classifier 19 may be expanded further to recognize additional signal types besides the speech and noise outputs for level estimator 21. It should be noted that the operation of signal classifier 19 and level estimator 21 may be considered to be bootstrapped by virtue of conductor 22 which feeds back a threshold signal (TN) from level estimator 21 to signal classifier 19. The two other outputs of level estimator 21 are the magnitude ~TM) decision level and the lower energy (TE) decision level~
MAY~ C J. JR. 16 '7 ~
whiCh are utilized by respective comparers 23 and 24 in speech definer 16. Por the other speech definer signal inputs, the single sample magnitude is applied to comparer 23 while the EMP signal is applied to comparer 24.
Comparer 23 serves to provide a rapid response to a loud talker while comparer 24 provides a slower, but sure response to a lower level talker. Reference 26 produces a fixed level for signal classifier 19 and comparer 27 of speech definer 16. The fixed level (FL) is used so that comparer 27 activates the speech detector for the typical -30 dBmO tone levels used for signaling purposes. This level corresponds in the ~ law coding format to level 32.
It should be pointed out that this feature is primarily a function of the system where the speech detector is used and, of course, may be modified or omitted without affecting the actual speech detection function. The outputs of comparers 23, 24 and 27 are logically combined in OR gate 28 to produce an output from the gate for speech definer 16 when the decision level of any comparer is exceeded.
After the output of gate 28 has ceased for a period of time, hangover timer 29 will have produced an output for the duration of the output of gate 28 extended by the hangover interval. Should the output of gate 28 reappear before the hangover interval lapses, hangover timer 29 will be reset and provide the bridging function which was previously mentioned. In any case, timer 29 will always extend the duration of the output of gate 28 by the prescribed duration of the hangover interval.
FIGS. 2 and 3 respectively depict the circuitry of signal classifier 19 and level estimator 21 of variable threshold generator 14 which develops dynamically adjustable decision levels for speech definer 16. Since this portion of the speech detector provides the basis of its operation and performance, generator 14 will be described in detail. Briefly, generator 14 is designed to take into account the particular talker and noise level on ~: :
MAY, C. J. JR. 16 the trunk currently being serviced to provide the best grade of service Eor all talkers within a relatively wide range of possible background noise levels and without producing excessive spurious activity indications at output 12. The specifics of the description to follow is based on the application dependent assumption that the digital encoding of the signals on trunk 11 is in accordance with the standard Unites States 8-bit ~ 255 format ( ~ law) wherein approximately 256 levels are used and the minimum negative and positive maximum levels of 127 are attained by the peaks of a +3 dBmO sinusoidal waveform.
However, those skilled in the art may readily apply the inventive principles to any desired encoding format and also to operate directly with analog signals.
In FIG. 2, the EMP average representation is applied to peak catcher and store 31 and minimum catcher and store 32 which serve respectively to extract and retain the peak (P) and minimum (M) of the EMP signal values.
Since the timing used to obtain these signal values is determinative on the outcome, controller 33 issues pulses for resetting stores 31 and 32 at appropriate times. For the time being, it is sufficient to describe controller 33 as being essentially a four state sequence machine which establishes appropriate time constants for each signal measure to match the variety of signaling conditions produced by a typical telecommunications trunk. These states are designated as IDLE, WAIT, NOISE ACTIVE and SPEECH/TONE ACTIVE. Each state has an associated conductor for indicating machine status to timing generator 34. The inhibit input to controller 33, which serves to suspend operation, will not be discussed further at this time since its function is used in the application depicted by FIG. 6.
Timing generator 34 serves as an interval timer which helps to control the state transitions that occur within controller 33. Additional information pertaining to controller 33 will be provided in the discussion of the state sequence diagram shown in FIG. 4. The EMP signal is MAY,C. J. JR. 16 also applied to comparer 36 which will produce a signal to indicate to controller 33 when the EMP signal is greater than a noise level of TN which is fed back from level estimator 21. The remaining signal input to controller 33 is provided by comparer 37. Each time the peak from store 31 exceeds the fixed level (FL) from reference 26 comparer 37 indicates same for controller 33. Comparers 38 and 39 each receive both the P and M signal values from stores 31 and 32. Comparer 38 provides a speech (SP) output to indicate a speech produced EMP signal when P>2M+16. Comparer 39, on the other hand, identifies a noise produced EMP signal with a NSE output when P<1.5M.
It is again noted that these particular ratios are a function of ~ law coding while the underlying principles are applicable to other coding formats.
FIG. 4 is a state diagram for describing the operation of controller 33 of FIG. 2. In other words, controller 33 is a particular one of what is known in the art as sequential machines which are usually described by way of state diagrams. See, for example, Donald R. Haring, Sequential-Circuit Synthesis: State Assignment Aspects (Cambridge, Mass.: The MIT Press 1966). As can be readily observed, there are the four previously identified states in FIG. 4. At the outset, it is to be understood that the operation of the circuitry of FIGS. 2 and 3 is associated with a particular one of the several trunks typically connected to a TASI System. Accordingly, these four states may be thought of as trunk status store states which each have individual time constants for associated signal measures that may be thought of as trunk signal history formed from accumulations of previous signal measures.
Further more, it is to be understood that the functioning of variable generator 14 is independent of the actual speech detection function provided by speech definer 16 of FIG. 1. However, generator 14 provides the variable dynamic threshold levels that the latter uses in speech detection.
. ,." .i ;:
,, ~ ` ' .
MAY, C. J. JR. 16 Ll~17~
Initially, controller 33 starts in the IDLE state where a counter within timing generator 34 (timer 1, T01) provides pulses to define intervals of 256 ms wherein the peak, P, and the minimum, M, of the E~P of signal energy presumed to be indicative of noise are directly measured and stored to characterized this noise. Also, the value of L, the long term minimum of M, may be incremented one unit in 1/4 second intervals in the direction of L = M should this be necessary. Another function which occurs in the IDLE state and between connections of the trunk is that K, the talker level estimate, is incremented one unit every 16 seconds to make it equal to a typical talker within a rather large range (~45dB.) defined by upper and lower limits. Each unit change is equivalent to a 3 dB level shift. In particular, high thresholds left over from loud talkers are reduced while low thresholds left over from weak talkers are increased in anticipation of a new talker at a more likely talker level. This operation of characterizing noise only takes place until the EMP level exceeds the TN level when controller 33 switches to the WAIT
state.
In the WAIT state, the first event which may possibly occur is that P of the EMP energy average exceeds a high fixed threshold of ~ law level 32 which corresponds to about -30 dBmO. Since this high energy level is assumed not to be noise, controller 33 will switch to the TALKER
ACTIVE state. If this does not occur, the noise from the trunk which produced the state transition most likely was a transient condition, such as a pop or a click, then timer l(T01) will time out in 64 ms and switch controller 33 to the NOISE ACTIVE state. However, the values of P and M obtained from this signal condition that produced this transition are stored and used to adjust L.
While in NOISE ACTIVE, another timer in timing generator 34 (timer 2) is triggered by signal transitions to produce a pulse repetition rate referred to as granularity, (GX) which is divided into two slower pulse rates designated as .:
MAY, C. J. JR. 16 1 ~ 5~
granularity even, (GE), and granu:Larity odd, (GO~. As illustrated in FIG. 5, the first pulse to occur of the two slower pulse rates after the signal transition is the GE
pulse. The letter S designates the spacing or timing interval of the pulse repetition rate. It will become evident that a variety of pulse repetition rates are utilized to provide different time constants appropriate to the signal measure being performed during the several signaling conditions. In each case, the designated timing serves to identify the appropriate pulse repetition rate.
Now if the noise signal previously referred to as a click or pop was produced by the trunk being switched for a new call with a higher background noise level, the previously experienced M and P will be reset at the occurrence of GE and GO pulses respectively in 512 ms intervals while controller 33 is in the NOISE ACTIVE state.
Also the GC pulses will serve for updating L in 1/~ second intervals. After the levels of M and P are used to adjust L and 256 ms elapse without the signal energy exceeding the TN noise threshold, controller 33 will switch back to the initial IDLE state. The 256 ms interval is provided by timer 1 to provide a hangover or bridging function. While in the NOISE ACTIVE state the ratio of P/M is calculated to ascertain the occurrence of a low level speech signal. If the ratio of P/M is high enough to indicate speech, i.e., P>2M + 16 or the value of P exceeds the fixed level here taken as ~ law level 32, controller 33 will switch to the TALKER ACTIVE state. The state path including IDLE, WAIT, NOISE ACTIVE and returning to IDLE, however, may typically occur a number of times without entering the TALKER ACTIVE
state due to various types of activities within the telephone plant. However, after following this path, the variable thresholds of Tl~ and TE are better reference levels for use in detecting speech activity.
When controller 33 switches to the TALKER ACTIVE
state from either the WAIT or NOISE ACTIVE states, timer 2 of timing generator 34 provides longer duration timing .
, .
MAY, C. J. JR. 16 intervals for the measures associated with speech activity.
These include resetting M and P upon the occurrence of respective granularity pulses GE and GO at 2 second intervals. This state is given a certain latitude of stability by timer 1, which provides 256 ms of hangover by resetting timer 1 any time the signal energy exceeds the TN
noise threshold during the hangover interval. Also, K is updated at one second intervals defined by the GX pulses and L is updated at two second intervals defined by the GE
pulse train. If the signal happens to be a tone, the P/M
ratio will be lower than 1.5 and no K update occurs. In addition, if the level of the pack, P, is less than the fixed threshold, and the signal is classified as not speech, a state transition results to the NOISE ACTIVE
state upon the occurrence of -the next GX pulse. If the signal level drops below the noise threshold levels for a period of 256 ms, controller 33 will switch back to the IDLE state and reset the values of P and M obtained from the current noise levels.
In FIG. 3, level estimator 21 receives the outputs from FIG. 2 designated P, M, SP, and NSE. In addition, there are outputs directly from controller 33 which are used for updating. These outputs are T update noise or TUL and T update speech or TUK. In level estimator 21, comparer 41 receives the M signal and the noise level L signal which generally corresponds to the long term minimum of EMP. The value of L is obtained from noise level store or NL store 43 which is incremented by noise control 42 when it is clocked at 1/4 second intervals via the TUL line from controller 33. From the comparison of M and L at the end of each update interval the value of L in store 43 may not be adjusted at all or adjusted one unit in either a positive or negative manner in the direction to obtain L = M. l`he other input to noise control serves as an enabling signal. l'his is the noise (NSE) signal obtained frorn comparer 39 of FIG. 2 which signals the presence of a noise signal when there is a low MAY, C J. JR. 16 ratio in the comparison of the current peak (P) to minimum (M) signal values. Each adjustment incrernent is equivalent to approximately a 1 dB change in the value of L retained by store 43. I`his stored value is then altered by translator 44 to form the noise threshold or TN signal value before application to adder 46. This alteration or adjustment is primarily a function of the coding characteristic used to form the digital signal in trunk 11.
For ~ law, translator 44 multiplies L by 1.5 and then adds one to form T~.
For the speech signal levels comparer 47, talk volume control 48, K store 49, and translator 51 perform a very similar operation. Since in this case the output of adder 46, designated TM, is the sum of the outputs from translators 44 and 51 and provides one input to comparer 47, the comparison must be again related to the properties of the coding format used in providing the digital signal. For a completely linear coding format, TM
would be compared with P times a factor slightly greater than one. For a strictly logarithmic coding format, TM
would be compared with P minus a constant dB value. In the case of ~ law, a piecewise comparison is made since low levels are linearly related while high levels are linear on a logarithmic scale. When 2P is less than forty-eight, the value of K is eventually decremented to zero at which point TM corresponds solely to the noise level. In the range where 2P is greater than forty-eight but less then eighty, T~ is compared with 2P-16. Then the value of K is incremented when TM is smaller and decremented when TN is compared to 2P-16. When P is larger than eighty, TM is compared to 2P-24 and the value of K will be increased when TM is smaller and decreased when T~l is larger than 2P-24.
The output of comparer 47 may, as previously stated, signal no adjustment as a result of these comparisons.
Another input to control 48 is the SP signal of comparer 38 which enables the adjustment only when the P/M
ratio is indicative of speech. Each adjustment increment MAY, C. J JR. 16 'Il~l'i'~
to the value in K store 49, is equivalent to about a 3 dB
change in value at the output of translator 51 which provides a level, designated as TS/N, since it is a ratio of the talker level to the noise level. For the ~ law coding characteristic, translator 51 actually performs a multiplication by a factor of eight.
Accordingly, the latter circuitry forms an estimate of talker level derived from the peak of the average representation (P) normalized by L the noise level estimate which is combined in adder 46 whose output is 'r~.
T~ is compared by comparer 23 of FIG. 1 directly with the single sample magnitude. T~ is also applied to translator 52 whose output is TE a lower decision level for comparison to the EMP (signal power or energy) by comparer 24. This latter comparison insures a sure response to lower level talking signals while the former comparison provides a fast response to higher level talking signals.
For ~ law, translator 52 divides by a factor of four.
FIG. 6 depicts the application of the speech detector embodiment of FIG. 1 in a typical two-way transmission environment subject to echo in the form of signal leakage on transmit trunk 111 from receive trunk 115. The basic speech detector for transmit trunk 111 comprises measure generator 113, variable threshold generator 114, transmit speech definer 116 and reference 126 which function to provide an activity signal on conductor 112 indicative of the occurrence of speech for echo protector 131. Incidentally, the reference numerals in FIG. 6 whose last two digits correspond to FIG. 1 are used to designate identical components.
It should be pointed out that in system applications, such as Time Assignment Speech Interpolation (TASI) or Digital Speech Interpolation (DSI), the value of L (noise level estimate) and the value of ~M (talker level estimate) are useful measures. For example, the value of I
can be used to provide noise fill to prevent the perception of disconnects in a TASI arrangement. The remaining . ' MAY, C. J. JR. 16 components in E`IG. 6, namely, measure generator 132, stretcher 133, receive speech definer 134, break-in definer 136 along with echo protector 131 serve as a receive speech detector for interrupting the operation of the transmit speech detector. This enables the correct designation of intervals for accurate decision level determination. In other words, adaption of decision level should only occur using transmit signals not contaminated by echo leakage signals. Furthermore, the ability to detect expected echo energy prevents a request for trunk service if only echo is present and also allows the transmit speech detector to request trunk service in spite of the presence of echo. It should also be pointed out that the arrangement of FIG. 6 is required even though some form of echo protection is used in the transmission environment since echo protection techniques typically will not always eliminate echoes.
Aside from the basic operation of the transmit speech detector which remains primarily the same as before, the differences in operation attributed to the components used in the receive portion of the arrangement of FIG. 6 will be described. Measure generator 132 provides a receive exponentially mapped past (RRMP~ signal derived from receive trunk 115. Stretcher 133 uses this signal at its input and a version of it internally delayed by about 24 ms to provide a stretched REMP(SREMP) signal output whose amplitude at any instant corresponds to the greater of current REMP signal and any other previous REMP signal in the designated delay interval. Another and, perhaps, more descriptive term for stretcher 133 is an echo envelope generator. The delay of stretcher 133 takes into account the signal propagation experienced as the receive signal travels over the leakage path (indicated by the dashed arrow in FIG. 6) of a remotely located hybrid transfor-ner where the signals of trunks 111 and 115 are combined for two wire transmission. This is the signal path which must be traveled by the receive signal before it appears on MAY, C. J. JR. 16 transmit trunk 111. Break-in definer 136 compares the TEMP
and SREMP signal levels obtained from respective measure generator 113 and stretcher 133. Definer 136 issues a signal to indicate that the transmit energy exceeds the stretched and delayed receive energy. The break-in signal is supplied to echo protector 133 along with the output of speech definer 116.
I'he output of measure generator 132 is also applied to speech definer 134 whose output indicates when the input exceeds a fixed threshold representative of minimal but significant signal energy. Thus the speech detector arrangement for the receive signal need only be a simple energy detector which serves to indicate the presence of significant signal energy whether it constitutes speech and/or tone. This signal is applied to the inhibit input of generator 114 to stop the adaptation of the decision levels when the signal levels of trunk 111 are contaminated by echo signal energy.
Echo protector 131 provides an output designated as trunk needs service (TNS) to indicate the occurrence of active speech signals on trunk 111. This output is produced when the signal on conductor 112 indicates speech signal activity and definer 136 produces a break-in (BI) signal. If the activity signal ceases, or BI ceases then the TNS signal switches to a level indicative of an idle condition. When the TEMP>SREMP ceases, the BI signal will switch to the idle state only after the termination of a hangover interval or speech activity ceases. A hangover duration on the order of 256 ms has proven to provide satisfactory performance.
FIG. 7 depicts the circuit diagram of stretcher 133 in FIG. 6. The vertical dashed lines in FIG. 7 partition the circuit into an input section which is a peak catcher, a middle section which saves the last A peaks, and an output section which selects the greater peak from the other two sections. To input 141, delay, ~ , 142 and comparer, C, 143 are connected for application of the REMP
MAY, C. J. JR. 16 11~81 ~
Signal. Delay 142 has an enable terminal connected for activation by the output of OR gate 144. Gate 144 forces delay 142 to accept the REMP signal each time a Gl pulse occurs. FIG. 8 illustrates the GT pulse and its relationship to the other pulses used to operate the circuit of FIG. 7. rrhe output of delay 142 is connected to the other input of comparer 143. Comparer 143 provides an output signal to enable delay 142 via gate 144 when the level of input REMP signal exceeds the signal level stored in delay 142. This condition is indicated by the expression I~>NEW assigned to the output of comparer 143 in FIG. 7.
The function of the input section is to follow the rising peaks or slopes of REMP signal and then to retain the maximum peak value for the duration of the ~ interval. The increasing signal level is provided access to delay 142 by virtue of the output of comparer 143. In particular, comparer 143 produces this output signal when the signal level at input 141 exceeds the NEW level provided by the present signal value stored in delay 142. When this condition terminates, delay 142 holds the peak until the next G ~ pulse occurs.
l'he middle section of the circuit of FIG. 7 includes delay line 146 which has A-l units of delay and a signal comparison portion which includes the circuitry between selector 147 and selector 148. In this circuitry, two signal paths are provided. One signal path extends from the output of selector 147 via delay 149 to a first signal input of selector 148. The other signal path is line 151 extending from the output of selector 147 directly to the other signal input of selector 148. Selector 147 has a control input to which is applied the GT pulse. In selector 147, the application of the Gl pulse completes the signal path from the output of delay 142 to the input of delay 149. Similarlyr the occurrence of the high level signal, TA, at tne control input of selector 148 via OR
gate 153 completes a conductive path from the output of ;
MAY~ C. J. JR. 16 delay l4g to the input of delay line 146. This transfers the contents QA of delay 149 to the Ql unit of delay line 146 but not to delay 152 since Gl is absent. When the control signal is low, the other signal input of selectors 147 and 148 is coupled to their respective outputs.
As can be observed from E`IG. 7, the control input of selector 148 is connected to the output of OR gate 153 and OR gate 154 is associated with the enable input of delay 149. Accordingly, the occurrence of ~A pulse not only enables delay 149 to accept its signal input but also makes its output available at the output of selector 148 for delay line 146. The other input of gate 153 is connected to comparer 156 which enables the gate when the output level of delay 149 is greater than the signal level on line 151.
The purpose of the middle section of the circuit of FIG. 7 is to store the A past I interval peaks. During the G~ pulse the input to delay 149 is the NEW signal from the output of delay 142 while the output of selector 148 is the larger of the value retained in delay 149 or the NEW
signal on line 151 due to the operation of comparer 156.
During the portion of the pulse Tl-Ta not coextensive with the TA pulse, the input to delay 149 is QOLD, the previously saved I interval peaks stored in delay 146. Now the output of delay 149 is the previously stored NEW value since the output of gate 154 is not enabling the input to the delay at this time. Accordingly, the output of selector 148 is the larger of the old stored values or NEW
which is also due to the operation of comparer 156. When the TA pulse occurs, the contents of delay 149 is applied to the first unit of delay line 146 while delay 149 is being loaded with the contents of the last unit in delay line 146 which is the oldest of the previous stored values.
This process will be repeated upon occurrence of the next Gl pulse.
MAY, C. J. JR. 16 The output section of the circuitry of FIG. 7 includes previously mentioned delay 152 which provides a signal input to comparer 157 and selector 158.
Attenuator 159 is connected to the output of selector 158 to provide the SREMP signal with a loss adequate to compensate for the minimum expected return loss of the echo path symbolically indicated in FIG. 6. For TASI systems, this loss is about 6 db. The other signal inputs to comparer 157 and selector 158 are provided by the output of delay 142 which is the NEW signal. Comparer 157 provides the control input for selector 158 so that the selector's output always corresponds to the larger of the NE~ Signal or the OLD signal.
For a typical application in a TASI system specific values will provide satisfactory performance. For example, the precision may be defined in terms of time accuracy, l=4 ms, and bit accuracy, N=8 bits. A total delay of 24 ms is provided with A=4 units in delay line 146 along with the presence of delays 142 and 152 in the input and output sections of stretcher 133. Of course, in other applications different specific values may be readily selected by those working in the art in the art to provide suitable performance.
What has been described hereinabove is but illustrative of the principles of my invention. For example, it will be appreciated that applicant~s speech detector, though shown and described in connection with a single transmission channel for purposes of describing the invention, may be shared advantageously among a plurality of transmission channels on a time division multiplexed basis, each channel being connected to the speech detector during a distinct time slot. Numerous and varied other arrangements may be derived by those skilled in the art without departing from the spirit and scope of the invention.
Claims (9)
1. Apparatus for indicating the occurrence of speech in a signal indicative of both speech and noise, the apparatus including:
means for generating a representation of the average magnitude of the signal during a moving time interval;
the apparatus being characterized by classifying means for receiving said representation and a noise level estimate, said classifying means generating a first output to indicate when said representation has prescribed attributes indicative of speech and as second output to indicate when said representation has prescribed attributes indicative of noise;
level estimator means responsive to said first and second outputs and said representation, said level estimator means providing a noise level estimate using the portion of said representation identified by the occurrence of said second second output, said level estimator means providing a first decision level output by combining said noise level estimate and the portion of said representation defined by the occurrence of said first output in excess of a prescribed amount of said first decision level output; and comparing means for providing an output indicative of the occurrence of speech signal activity when said first decision level is exceeded by the signal.
means for generating a representation of the average magnitude of the signal during a moving time interval;
the apparatus being characterized by classifying means for receiving said representation and a noise level estimate, said classifying means generating a first output to indicate when said representation has prescribed attributes indicative of speech and as second output to indicate when said representation has prescribed attributes indicative of noise;
level estimator means responsive to said first and second outputs and said representation, said level estimator means providing a noise level estimate using the portion of said representation identified by the occurrence of said second second output, said level estimator means providing a first decision level output by combining said noise level estimate and the portion of said representation defined by the occurrence of said first output in excess of a prescribed amount of said first decision level output; and comparing means for providing an output indicative of the occurrence of speech signal activity when said first decision level is exceeded by the signal.
2. Apparatus in accordance with claim 1 wherein said level estimator means comprises translating means for using the first decision level as an input to provide a second decision level of a prescribed lower magnitude than said first decision level, and said comparing means uses the second decision level and said representation for providing the output indicative of the occurrence of speech signal activity when the second decision level is exceeded by said representation.
MAY, C. J. JR. 16
MAY, C. J. JR. 16
3. Apparatus according to claim 2 wherein said classifying means comprises:
measuring means for obtaining two values of the representation, a first value corresponding to the peak and the second value corresponding to the minimum of the representation, and comparing means, connected to receive the two values, for providing the second output when the first value is less than a prescribed constant times the second value.
measuring means for obtaining two values of the representation, a first value corresponding to the peak and the second value corresponding to the minimum of the representation, and comparing means, connected to receive the two values, for providing the second output when the first value is less than a prescribed constant times the second value.
4. Apparatus according to claim 3 wherein said level estimator means comprises:
comparing means for comparing the second value to a long term minimum value and producing an output indicative of the relative magnitude of the values, storing means for retaining the long term minimum value, and means for altering the contents of said storing means in response to the output indication of the comparing means by increasing the contents a prescribed amount when the minimum value is greater than the long term minimum value and by decreasing the contents a prescribed amount when the minimum value is less than the long term minimum value.
comparing means for comparing the second value to a long term minimum value and producing an output indicative of the relative magnitude of the values, storing means for retaining the long term minimum value, and means for altering the contents of said storing means in response to the output indication of the comparing means by increasing the contents a prescribed amount when the minimum value is greater than the long term minimum value and by decreasing the contents a prescribed amount when the minimum value is less than the long term minimum value.
5. Apparatus according to claim 3 wherein said classifying means further comprises:
comparing means connected to receive the two values and providing the first output when the first value is greater than twice the second value by a prescribed constant.
comparing means connected to receive the two values and providing the first output when the first value is greater than twice the second value by a prescribed constant.
6. Apparatus according to claim 5 wherein the signal is sampled and digitally encoded and the level estimator means further comprises:
translating means connected to receive the long term minimum value and providing a noise level compatible with the encoded format used to form the digitally encoded signal samples.
translating means connected to receive the long term minimum value and providing a noise level compatible with the encoded format used to form the digitally encoded signal samples.
7. Apparatus according to claim 6 wherein said level estimator means further comprises:
comparing means connected to receive the first value and the first decision level for producing an output indicative of the relative magnitude of the values in the comparison;
adjusting means for producing a positive increment when the peak indicates an increase in talker volume;
said adjusting means producing a negative increment when peak value indicates a decrease in talker volume;
storing means for receiving the increments, said storing means accumulating and retaining the total of the increments to provide a level indicative of talker level relative to noise level, and adding means for combining the noise level from the translating means with the talker level from the storing means to provide the first decision level.
comparing means connected to receive the first value and the first decision level for producing an output indicative of the relative magnitude of the values in the comparison;
adjusting means for producing a positive increment when the peak indicates an increase in talker volume;
said adjusting means producing a negative increment when peak value indicates a decrease in talker volume;
storing means for receiving the increments, said storing means accumulating and retaining the total of the increments to provide a level indicative of talker level relative to noise level, and adding means for combining the noise level from the translating means with the talker level from the storing means to provide the first decision level.
8. An arrangement for detecting speech signal activity in transmission signals indicative of speech and noise, the transmission signal traversing in a first direction, the arrangement comprising;
first means for producing a representation of the transmission signals by weighted averaging the signal occurring over a predetermined recent interval of time;
second means, connected to the first means, for producing and maintaining an output indicative of a peak value of the representation;
third means, connected to the first means, for producing and maintaining an output indicative of a minimum value of the representation;
controlling means, including transition means capable of assuming a prescribed plurality of states occurring in a sequence responsive to predetermined signal conditions, for resetting the second and third means at different intervals according to each one of the prescribed MAY, C. J. JR. 16 - 22 -plurality of states;
fourth means, in circuit with outputs of the second and third means, for indicating signal activity characteristic of speech when the outputs relate to each other within a first predetermined ratio range;
fifth means, in circuit with the outputs of the second and third means, for indicating signal activity characteristic of noise when the outputs relate to each other within a second predetermined ratio range exclusive of the first predetermined ratio range;
noise level estimating means, connected to the third means and the fifth means, for comparing a stored noise level to the minimum value and altering the stored value prescribed amounts in the direction to achieve equality at intervals defined by an updating signal, the controlling means producing the updating signal after state transitions from active states and while in active states at a predetermined rate;
talker level estimator means, connected to the second means and the fourth means, for comparing a stored talker level estimate with a sum indicative of the noise level estimate and the current talker level estimate and changing the stored value a prescribed amount upon the occurrence of a second updating signal and when necessary to achieve a more accurate representation of the actual talker level, the controlling means producing the second updating signal while in an idle state and the talker active state at prescribed times; and speech defining means for receiving the representation of the transmission signals and the sum to provide an activity signal indicative of speech when the representation exceeds the sum.
first means for producing a representation of the transmission signals by weighted averaging the signal occurring over a predetermined recent interval of time;
second means, connected to the first means, for producing and maintaining an output indicative of a peak value of the representation;
third means, connected to the first means, for producing and maintaining an output indicative of a minimum value of the representation;
controlling means, including transition means capable of assuming a prescribed plurality of states occurring in a sequence responsive to predetermined signal conditions, for resetting the second and third means at different intervals according to each one of the prescribed MAY, C. J. JR. 16 - 22 -plurality of states;
fourth means, in circuit with outputs of the second and third means, for indicating signal activity characteristic of speech when the outputs relate to each other within a first predetermined ratio range;
fifth means, in circuit with the outputs of the second and third means, for indicating signal activity characteristic of noise when the outputs relate to each other within a second predetermined ratio range exclusive of the first predetermined ratio range;
noise level estimating means, connected to the third means and the fifth means, for comparing a stored noise level to the minimum value and altering the stored value prescribed amounts in the direction to achieve equality at intervals defined by an updating signal, the controlling means producing the updating signal after state transitions from active states and while in active states at a predetermined rate;
talker level estimator means, connected to the second means and the fourth means, for comparing a stored talker level estimate with a sum indicative of the noise level estimate and the current talker level estimate and changing the stored value a prescribed amount upon the occurrence of a second updating signal and when necessary to achieve a more accurate representation of the actual talker level, the controlling means producing the second updating signal while in an idle state and the talker active state at prescribed times; and speech defining means for receiving the representation of the transmission signals and the sum to provide an activity signal indicative of speech when the representation exceeds the sum.
9. The arrangement of claim 8 for detecting speech signal activity including means for protecting from echo derived from a second transmission signal traversing in a second direction, the controlling means having an inhibit input capable of inhibiting its operation, and the means for protecting comprising;
sixth means for producing an average representation of the second transmission signal;
seventh means connected to receive the representation from the sixth means and producing an output for the inhibit input of the controlling means to inhibit its operation when the representation is greater than a predetermined level;
eighth means for producing an output indicative of the greater of the peak of the average representation or the peak of an internally delayed version of the average representation;
ninth means for comparing the output of the eighth means to the representation produced by the first means and producing an output to indicate when the latter is greater; and tenth means connected to receive the activity signal and the output of the ninth means and providing an output requesting service during the occurrence of the activity signal and the output from the ninth means.
sixth means for producing an average representation of the second transmission signal;
seventh means connected to receive the representation from the sixth means and producing an output for the inhibit input of the controlling means to inhibit its operation when the representation is greater than a predetermined level;
eighth means for producing an output indicative of the greater of the peak of the average representation or the peak of an internally delayed version of the average representation;
ninth means for comparing the output of the eighth means to the representation produced by the first means and producing an output to indicate when the latter is greater; and tenth means connected to receive the activity signal and the output of the ninth means and providing an output requesting service during the occurrence of the activity signal and the output from the ninth means.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA000429985A CA1164351A (en) | 1980-01-25 | 1983-06-08 | Multiple variable threshold speech detector |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US06/115,185 US4277645A (en) | 1980-01-25 | 1980-01-25 | Multiple variable threshold speech detector |
| US115,185 | 1980-01-25 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA1158174A true CA1158174A (en) | 1983-12-06 |
Family
ID=22359780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA000368374A Expired CA1158174A (en) | 1980-01-25 | 1981-01-13 | Multiple variable threshold speech detector |
Country Status (8)
| Country | Link |
|---|---|
| US (1) | US4277645A (en) |
| JP (1) | JPS56142600A (en) |
| CA (1) | CA1158174A (en) |
| DE (1) | DE3101775A1 (en) |
| FR (2) | FR2482389B1 (en) |
| GB (2) | GB2068698B (en) |
| IT (1) | IT1135165B (en) |
| NL (1) | NL8100323A (en) |
Families Citing this family (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4357491A (en) * | 1980-09-16 | 1982-11-02 | Northern Telecom Limited | Method of and apparatus for detecting speech in a voice channel signal |
| US4376874A (en) * | 1980-12-15 | 1983-03-15 | Sperry Corporation | Real time speech compaction/relay with silence detection |
| USRE32172E (en) * | 1980-12-19 | 1986-06-03 | At&T Bell Laboratories | Endpoint detector |
| US4370521A (en) * | 1980-12-19 | 1983-01-25 | Bell Telephone Laboratories, Incorporated | Endpoint detector |
| WO1982003144A1 (en) * | 1981-03-05 | 1982-09-16 | Western Electric Co | Energy band discriminator |
| US4410763A (en) * | 1981-06-09 | 1983-10-18 | Northern Telecom Limited | Speech detector |
| JPS59139099A (en) * | 1983-01-31 | 1984-08-09 | 株式会社東芝 | Voice section detector |
| US4535445A (en) * | 1983-06-16 | 1985-08-13 | At&T Information Systems | Conferencing system adaptive signal conditioner |
| US4920534A (en) | 1986-02-28 | 1990-04-24 | At&T Bell Laboratories | System for controllably eliminating bits from packet information field based on indicator in header and amount of data in packet buffer |
| US4829578A (en) * | 1986-10-02 | 1989-05-09 | Dragon Systems, Inc. | Speech detection and recognition apparatus for use with background noise of varying levels |
| US5008941A (en) * | 1989-03-31 | 1991-04-16 | Kurzweil Applied Intelligence, Inc. | Method and apparatus for automatically updating estimates of undesirable components of the speech signal in a speech recognition system |
| US4979214A (en) * | 1989-05-15 | 1990-12-18 | Dialogic Corporation | Method and apparatus for identifying speech in telephone signals |
| DK406189A (en) * | 1989-08-18 | 1991-02-19 | Otwidan Aps Forenede Danske Ho | METHOD AND APPARATUS FOR CLASSIFYING A MIXED SPEECH AND NOISE SIGNAL |
| GB2256351B (en) * | 1991-05-25 | 1995-07-05 | Motorola Inc | Enhancement of echo return loss |
| US5471527A (en) | 1993-12-02 | 1995-11-28 | Dsc Communications Corporation | Voice enhancement system and method |
| JP3483695B2 (en) * | 1996-03-14 | 2004-01-06 | 株式会社リコー | Voice communication device |
| US5884255A (en) * | 1996-07-16 | 1999-03-16 | Coherent Communications Systems Corp. | Speech detection system employing multiple determinants |
| FR2768544B1 (en) * | 1997-09-18 | 1999-11-19 | Matra Communication | VOICE ACTIVITY DETECTION METHOD |
| USD419160S (en) * | 1998-05-14 | 2000-01-18 | Northrop Grumman Corporation | Personal communications unit docking station |
| USD421002S (en) * | 1998-05-15 | 2000-02-22 | Northrop Grumman Corporation | Personal communications unit handset |
| US6223062B1 (en) | 1998-05-15 | 2001-04-24 | Northrop Grumann Corporation | Communications interface adapter |
| US6243573B1 (en) | 1998-05-15 | 2001-06-05 | Northrop Grumman Corporation | Personal communications system |
| US6169730B1 (en) | 1998-05-15 | 2001-01-02 | Northrop Grumman Corporation | Wireless communications protocol |
| US6041243A (en) * | 1998-05-15 | 2000-03-21 | Northrop Grumman Corporation | Personal communications unit |
| US6304559B1 (en) | 1998-05-15 | 2001-10-16 | Northrop Grumman Corporation | Wireless communications protocol |
| US6141426A (en) * | 1998-05-15 | 2000-10-31 | Northrop Grumman Corporation | Voice operated switch for use in high noise environments |
| DE19844702C2 (en) * | 1998-09-29 | 2002-11-21 | Siemens Ag | Method, receiving device and radio station for recognizing a user data block with incorrect user data symbols |
| US6314396B1 (en) * | 1998-11-06 | 2001-11-06 | International Business Machines Corporation | Automatic gain control in a speech recognition system |
| US6381570B2 (en) * | 1999-02-12 | 2002-04-30 | Telogy Networks, Inc. | Adaptive two-threshold method for discriminating noise from speech in a communication signal |
| US6490556B2 (en) * | 1999-05-28 | 2002-12-03 | Intel Corporation | Audio classifier for half duplex communication |
| US7263074B2 (en) * | 1999-12-09 | 2007-08-28 | Broadcom Corporation | Voice activity detection based on far-end and near-end statistics |
| JP4810044B2 (en) * | 2000-01-27 | 2011-11-09 | ニュアンス コミュニケーションズ オーストリア ゲーエムベーハー | Voice detection device with two switch-off criteria |
| DE10007752A1 (en) * | 2000-02-19 | 2001-08-23 | Braun Gmbh | Decision level setting method for analog to digital conversion of multi-level analog signal in digital receiver, involves using difference of low and high level signal, and high level signal as initial values and sampling signal |
| US20020103636A1 (en) * | 2001-01-26 | 2002-08-01 | Tucker Luke A. | Frequency-domain post-filtering voice-activity detector |
| US7146314B2 (en) * | 2001-12-20 | 2006-12-05 | Renesas Technology Corporation | Dynamic adjustment of noise separation in data handling, particularly voice activation |
| JP3812887B2 (en) * | 2001-12-21 | 2006-08-23 | 富士通株式会社 | Signal processing system and method |
| US7925510B2 (en) * | 2004-04-28 | 2011-04-12 | Nuance Communications, Inc. | Componentized voice server with selectable internal and external speech detectors |
| CN1299254C (en) * | 2004-05-31 | 2007-02-07 | 中兴通讯股份有限公司 | Approaching end voice detection realizing method for echo inhibitor |
| US20060241937A1 (en) * | 2005-04-21 | 2006-10-26 | Ma Changxue C | Method and apparatus for automatically discriminating information bearing audio segments and background noise audio segments |
| CN101859568B (en) * | 2009-04-10 | 2012-05-30 | 比亚迪股份有限公司 | Method and device for eliminating voice background noise |
| US8995823B2 (en) | 2012-07-17 | 2015-03-31 | HighlightCam, Inc. | Method and system for content relevance score determination |
| KR20140147587A (en) * | 2013-06-20 | 2014-12-30 | 한국전자통신연구원 | A method and apparatus to detect speech endpoint using weighted finite state transducer |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4008375A (en) * | 1975-08-21 | 1977-02-15 | Communications Satellite Corporation (Comsat) | Digital voice switch for single or multiple channel applications |
| US4029912A (en) * | 1975-12-10 | 1977-06-14 | Bell Telephone Laboratories, Incorporated | Common control digital echo suppressor |
| US4028496A (en) * | 1976-08-17 | 1977-06-07 | Bell Telephone Laboratories, Incorporated | Digital speech detector |
| US4167653A (en) * | 1977-04-15 | 1979-09-11 | Nippon Electric Company, Ltd. | Adaptive speech signal detector |
-
1980
- 1980-01-25 US US06/115,185 patent/US4277645A/en not_active Expired - Lifetime
-
1981
- 1981-01-13 CA CA000368374A patent/CA1158174A/en not_active Expired
- 1981-01-21 DE DE19813101775 patent/DE3101775A1/en not_active Withdrawn
- 1981-01-22 FR FR8101144A patent/FR2482389B1/en not_active Expired
- 1981-01-23 JP JP889181A patent/JPS56142600A/en active Granted
- 1981-01-23 NL NL8100323A patent/NL8100323A/en not_active Application Discontinuation
- 1981-01-23 IT IT19308/81A patent/IT1135165B/en active
- 1981-01-26 GB GB8102335A patent/GB2068698B/en not_active Expired
- 1981-05-18 FR FR8109843A patent/FR2478909B1/fr not_active Expired
-
1983
- 1983-12-02 GB GB08332290A patent/GB2136253B/en not_active Expired
Also Published As
| Publication number | Publication date |
|---|---|
| JPS56142600A (en) | 1981-11-06 |
| FR2478909B1 (en) | 1985-12-27 |
| GB2068698B (en) | 1984-10-24 |
| US4277645A (en) | 1981-07-07 |
| GB2068698A (en) | 1981-08-12 |
| FR2482389B1 (en) | 1985-12-27 |
| FR2482389A1 (en) | 1981-11-13 |
| GB2136253B (en) | 1985-05-09 |
| JPH0247142B2 (en) | 1990-10-18 |
| IT8119308A0 (en) | 1981-01-23 |
| GB2136253A (en) | 1984-09-12 |
| FR2478909A1 (en) | 1981-09-25 |
| GB8332290D0 (en) | 1984-01-11 |
| DE3101775A1 (en) | 1982-01-07 |
| IT1135165B (en) | 1986-08-20 |
| NL8100323A (en) | 1981-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA1158174A (en) | Multiple variable threshold speech detector | |
| AU596333B2 (en) | Technique for improved subjective performance in a communication system using attenuated noise-fill | |
| US4382164A (en) | Signal stretcher for envelope generator | |
| US4712235A (en) | Method and apparatus for improved control and time sharing of an echo canceller | |
| Bullington et al. | Engineering aspects of TASI | |
| US4028496A (en) | Digital speech detector | |
| US4644108A (en) | Adaptive sub-band echo suppressor | |
| EP0530423B1 (en) | Adaptive echo cancellation method and device for implementing said method | |
| CA1210541A (en) | Conferencing system adaptive signal conditioner | |
| EP0047590B1 (en) | Method of and apparatus for echo detection in voice channel signals | |
| US4059730A (en) | Apparatus for mitigating signal distortion and noise signal contrast in a communications system | |
| Miedema et al. | TASI quality—Effect of speech detectors and interpolation | |
| US4167653A (en) | Adaptive speech signal detector | |
| US4365112A (en) | Speech detector circuit for a TASI system | |
| Brady et al. | Echo suppressor design in telephone communications | |
| US4192979A (en) | Apparatus for controlling echo in communication systems utilizing a voice-activated switch | |
| GB2072466A (en) | Time assigned speech interpolation system | |
| CA1164351A (en) | Multiple variable threshold speech detector | |
| US6035033A (en) | Method and apparatus for limiting residual echo in a speech signal-carrying channel or line | |
| CA1159971A (en) | Speech detector circuit with associated gain control for a tasi system | |
| US4349707A (en) | System for measuring the attenuation on a transmission path | |
| CA1130920A (en) | Speech detector with variable threshold | |
| US5850437A (en) | Transmission system with improved tone detection | |
| US4360713A (en) | Echo detector particularly for speech interpolation communication systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| MKEX | Expiry |