WO1994000944A1

WO1994000944A1 - Method and apparatus for ringer detection

Info

Publication number: WO1994000944A1
Application number: PCT/US1993/006138
Authority: WO
Inventors: Brian L. Hinman; Shan-Shan Huang; Eric K. Gaut
Original assignee: Polycom, Inc.
Priority date: 1992-06-30
Filing date: 1993-06-28
Publication date: 1994-01-06

Abstract

The apparatus and method for detecting the presence of ringer signals in a telephone which samples the incoming telephone signals within a prescribed time interval, windows the signals, and computes a set of auto-correlation coefficients. These coefficients are normalized (58), and their absolute values (60) are calculated and summed (62) and compared with an empirically-derived threshold. If the coefficient sum exceeds the threshold (64), a high degree of correlation within incoming signals is indicated, distinguishing the incoming signal as containing ringer signal constituents.

Description

METHOD AND APPARATUS FOR RINGER DETECTION

CROSS-REFERENCE TO RELATED APPLICATIONS

The subject matter of this application is related to and incorporates by reference the subject matter of copending application Serial Number xxx,xxx, entitled "Method and Apparatus For Double-Talk Detector", filed June 12, 1992, by Shan-Shan Huang, Brian L. Hinman and Eric Gaut, and commonly assigned with the present invention.

BACKGROUND OF THE INVENTION

1. Field of the Invention

This invention relates to telephone technology and more particularly to the use of a ringer detector for identifying the presence of ringer signals during telephone call set-up.

2. Description of the Background Art

Traditional speaker phones function in a half-duplex mode, allowing only one person at a time to speak. When one talker (near-end) is speaking in a half-duplex system, the signals received from the other talker (far-end) are blocked until the near end speech is either completed or interrupted by a stronger signal. Often, while a talker is speaking, the signals may be blocked if someone at the other end makes moderate noises and activates the microphone. In other cases, talkers are required to shout into the speaker phone in order to be heard by the other end. It is generally very difficult, if not impossible, in a half-duplex system to interrupt current talkers while they are speaking. All of these conditions are very annoying in a teleconferencing situation.

Room acoustic echo has always been one of the most severe problems in hands-free speaker phone systems. Acoustic echoes occur when the far-end speech sent over the telephone line comes out from the near-end loudspeaker, feeds back into a nearby microphone, and then travels back to the originating site. Talkers at the far-end location can hear their own voices coming back slightly after they have just spoken. One method of eliminating these irritating acoustic echoes is to apply an echo suppresser to switch off the microphone while the other end is talking. This results in the half-duplex operation currently implemented on standard speaker phones. More sophisticated acoustic echo cancelers are available for full-duplex operation to improve interactivity in teleconferencing. Acoustic echo cancelers employ adaptive filtering techniques to model the impulse response of the conference room in order to reproduce the echoes from the speaker signal. The estimated echoes are then subtracted from the out-going microphone signals to prevent these echoes from going back to the far-end. The adaptive filter is intended to operate over a broad spectrum of audible frequencies. Because of this wide band of operation, updating the filter coefficients when the limited frequency components found in the ringer signal are received causes undesired skewing in the filter adaption. A reliable ringer detector is needed to accurately detect the presence of ringer signals in order to disable the adaptive process and prevent the filter from updating when ringer signals are present.

SUMMARY OF THE INVENTION

In accordance with the present invention, a method and apparatus are disclosed for detecting the presence of ringer signals in telephone communication. The method involves windowing a block of telephone signals received from the far-end. A set of auto-correlation coefficients are then normalized and summed to provide an indicator of the correlation within the incoming signal. Since ringer signals contain a small number of discrete frequencies, the correlation of the ringer signal will be high. The sum of the normalized auto-correlation coefficients is compared to an empirically- derived threshold and used to set a double-talk flag and thereby disable filter updating when ringer signals are present.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating major components of the preferred embodiment of a speaker phone incorporating a ringer detector of the present invention;

FIG. 2 illustrates a flow chart showing signal flows of the double-talk detector incorporated by the speaker phone of FIG. 1; and FIG. 3 contains a flow chart showing signal flows of an embodiment of the ringer detector of the present invention.

DESCRIPTION OF THE PREFERRED EMBODIMENT

Referring now to FIG. 1 there is shown a speaker phone 10 including a ringer detector 66 which detects the presence of ringer signals. Speaker phone 10 comprises two originating signal paths, including speaker signal 22 and microphone signal 24. The speaker signal 22 comes from the far-end via telephone line 44, propagate to and through the near-end speaker 42, and are

-2- heard by listeners in the near-end. Speech originating from the near-end is detected by microphone 40 and transmitted to the far-end along telephone lines 44. The microphone signal 24 coming from the near-end can also include echo coming from the far-end via speaker 42. The primary purpose of digitally processing these speaker and microphone signals (22, 24) is to remove the echoes in microphone signal 24 so that listeners in the far-end will only hear the near-end talkers, not their own voices coming back after the round- trip delay of telephone transmission. Signal processing also removes the echoes and side tones in speaker signal 22 so that only the far-end speech can be heard in the near-end.

Analog signals produced by talkers in the near-end are picked up by microphone 40, low pass filtered and digitized at 8 kilohertz into 16-bit integers by filter and A/D converter 11, which in the preferred embodiment may be implemented using a model 7525 codec commercially available from AT&T. Microphone signal 24 is passed through flattening filter 29 and transmitted to room double-talk detector 31. The purpose of flattening filter 29 is to compensate for any spectral variation introduced into microphone input signal 75 through the communication path traversed by processed speaker signal 51, including digital filter 33, filter and D/A converter 35, speaker 42, microphone 40, and filter and A/D converter 11. Room double-talk detector 31 receives processed speaker signals 51 and microphone input signals 75 from microphone 40 as inputs and determines whether speech originating in the near-end is present. If speech in the near-end is being generated, room double- talk detector 31 sets double-talk flag 85. If this flag is set (i.e., if near-end speech is being generated), the process of updating filter coefficients by Least Mean Square (LMS) filter 27 of conventional design is disabled to prevent the LMS filter coefficients from diverging. On the other hand, if double-talk flag 85 is not set, LMS filter 27 will adapt to keep track of changes in the surrounding environment, while estimating far-end echoes present in microphone signals 24.

Ringer detector 66 operates on processed speaker signal 51 in order to detect the presence of incoming ringer signals from the far-end. LMS filter 27 is intended to operate over a broad spectrum of audible frequencies, and updating the filter coefficients when the discrete frequency components found in the ringer signals are received, causes undesired skewing in the adaption of filter 27. Ringer detector 66 monitors incoming processed speaker signal 51 to determine whether ringer signals are present. When ringer signals are

-3- detected, double-talk flag 85 is set, thereby disabling the updating of LMS filter 27 coefficients.

Echoes arriving in microphone signal 24 are subtracted in summer 13 so that only signals generated in the near-end are sent to the far-end. Microphone signal 24 levels are adjusted by limiter 15 and microphone gain and attenuator 17. The adjusted microphone signal 36 is then used as a reference signal for Line Echo Canceler 19, and converted from digital to analog in block 21 before being transferred to telephone line 44.

Speaker signals 22 from telephone line 44 are digitized at 8 kilohertz by filter and A/D converter 23. The digitized speaker signals along with adjusted microphone signals 36 are used as inputs to line echo canceler 19. After passing through the line echo canceler (LEC) 19, the speaker signals are adjusted by limiter 26 and volume control and attenuator 25. This processed speaker signal 51 is then used as a reference signal for LMS filter 27, for the room double-talk detector 31, and for ringer detector 66 of the present invention. Before the processed speaker signal 51 is propagated through speaker 42 to the near end, the signal is flattened by digital filter 33 in order to compensate for any spectral variations introduced by speaker 42. Signal flattening involves the equalization of the frequency response to flatten the shape of the spectrum at microphone input 24. Conversion from digital to analog occurs in filter and D/A converter 35.

Speaker phone 10 employs a conventional adaptive LMS filter 27 for modeling the near-end room impulse response, and for simulating echoes from the processed speaker signal 51. These echoes are then subtracted from the near-end signal to prevent the echoes from going back to the far-end. LMS filter 27 adjusts its coefficients constantly to keep track of the changes in the near-end environment. However, coefficient update of filter 27 should be disabled when near-end speech or far-end ringer signals are present. Ringer detector 66 senses when ringer signals are present and disables the coefficient update of LMS filter 27 when ringer signals are detected.

Room double-talk detector 31 is based on standard techniques of Linear Predictive Coding (LPC). (See, for example, J.D. Markel, A.H. Gray, Jr., "Linear Prediction of Speech," Springer- Verlag, 1976.) Linear predictive coding is used to model the smoothed spectrum envelope of processed speaker signal 51 by an all-pole filter. The auto-correlation coefficients produced in room double- talk detector 31 are normalized, summed, and compared to an empirical threshold in ringer detector 66 in order to determine whether incoming ringer

-4- signals are present in processed speaker signal 51. If ringer signals are found, LMS filter 27 updating is disabled.

Referring now to FIG. 2, the detailed operation of double-talk detector 31 is shown in block diagram form. Processed speaker signal 51 is buffered over a period of 20 milliseconds for a total of 160 samples, and the energy of each 160- sample block is computed 53. This energy computation is first used to obtain the background noise level from telephone line 44. The background noise level is set dynamically and updated with each new block of samples. On start¬ up, the background noise level is set to some fixed, empirically determined value. If the energy level computed in 53 is less than the background noise level, the background noise level is reduced to the newly-measured energy level. If the energy level is greater than the background noise level, the background noise level is increased slightly with each such comparison, until it reaches a predetermined maximum value. This maximum value is necessary so that constant strong signals from the far-end will not be mistaken as transmission line noise. A threshold value is computed 53 as a function of the background noise level. If the comparison 55 determines that the energy value is below the threshold, then the double-talk flag 85 (indicating a silent condition) is set 65. Double-talk flag 85 is set, since it is desirable to disable the coefficient update when there are not any valid reference signals for LMS filter 27 (FIG. 1), even if the near-end is also silent. However, if the computed 53 energy is greater than the current threshold in 55, further analysis is required to determine the double-talk flag 85 status. If processed speaker signal 51 energy is greater than the threshold in 55, LPC techniques are then used to accurately detect the present of near-end signals.

If the computed energy exceeds the threshold in comparison 55, a conventional Hamming window 59 (see, for example, Oppenheim & Schafer, Digital Signal Processing. 1975, p. 242) is applied to the block of processed speaker signal 51 samples and a ten-pole LPC analysis 61 is performed using a standard auto-correlation method. A set of filter parameters 100 is generated by LPC analysis 61 and is transferred to filter and logic circuit 67 for double-talk detection. A set of 11 auto-correlation coefficients are transferred to ringer detector 66 (FIG. 3).

Referring now to FIG. 3, a flow diagram illustrates the detailed operation of the preferred ringer detector 66. LPC analysis 61 produces eleven auto-correlation coefficients 56: R(0), R(1)...R(10). These coefficients 56 are utilized by ringer detector 66 to determine whether processed speaker signal 51 contains ringer signals. Coefficients R(l)-R(10) are normalized 58 with respect

-5- to R(0), which is then set to unity. Since R(0) represents the energy of the signals and is always greater than or equal to each of R(l)-R(10), the maximum possible value which any of the remaining ten normalized coefficients R(l)- R(10) can have is unity. A unity value in each of the remaining coefficients would represent perfect correlation within the signals. By taking the absolute values 60 of each of the normalized coefficients and computing 62 the sum 63 of the normalized coefficients R(l)-R(10), a figure of merit is generated 63 as to the correlation of processed speaker signal 51. Since the ringer signal is highly correlated, generally comprising a limited set of discrete frequency sinusoids, the correlation of frequency components is high, and the ringer can be thus easily detected by comparing 64 the computed sum 63 to an empirically derived threshold. The preferred threshold used for this comparison is 5.0, although this number is not critical and may be varied widely to suit transmission and local telephone switching network standards. When the computed sum 63 exceeds the selected threshold, speaker signal 51 is determined to be sufficiently correlated to contain ringer signals.

Following the comparison 64 of the computed sum 63 to the threshold, an indicating signal 68 is generated and logically combined 65 with the result of the energy comparison 55, for the purpose of setting 65 and resetting double- talk flag signal 54. The result from LPC analysis 61 combines with the double- talk flag signal 54 to functionally control double-talk flag 85. Details of the operation of room double-talk detector 31 can be found in copending application Serial Number xxx,xxx, entitled "Method and Apparatus for Double-Talk Detector", filed June 12, 1992, by Shan-Shan Huang, Brian L. Hinman and Eric Gaut.

The double-talk flag 85 is then transmitted to LMS filter 27, and is used by the filter 27 to enable or disable coefficient updating. When double-talk flag 85 is not set, a non-ringer and near-end silent condition is indicated, and coefficient updating proceeds. When the double-talk flag is in a set condition, coefficient updating within LMS filter 27 is suspended until near-end or ringer signal origination ceases.

A method and apparatus have now been described which disclose the detection of ringer signals received during telephone communications. A set of eleven auto-correlation coefficients 56 is generated, and these coefficients 56 are normalized 58, the absolute values are taken 60 and the ten higher order normalized coefficients are summed 62 and compared to an empirically derived threshold. If the threshold value is exceeded, the received (processed speaker) signal 51 is recognized as being highly correlated and identified as a

-6- ringer signal. This is a preferred embodiment of the method and apparatus of the present invention. Variations of the above, such as rearrangement of steps and use of varying numbers of correlation coefficients, filter configurations and threshold values, will be clearly obvious to those ordinarily skilled in the art. It is therefore intended that the above invention only be limited by the claims appended below.

-7-

Claims

What is claimed is:

1. A method for detecting the presence of ringer signals in an incoming telephone signal, the method comprising the steps: analyzing the incoming telephone signal using a series of auto- correlation coefficients, the coefficients comprising a fundamental coefficient and a series of higher order coefficients; computing the sum of the auto-correlation coefficients which represents the correlation between frequency components comprising the incoming signal; and comparing the sum of auto-correlation coefficients to a predetermined threshold value in order to determine whether the incoming signal contains ringer signals.

2. The method according to claim 1, wherein the step of analyzing is preceded by windowing blocks of discrete time samples of the incoming telephone signal.

3. The method according to claim 2 of windowing, wherein the windowing of samples of the incoming telephone signal is performed using the Hamming window method.

4. The method according to claim 2 of windowing, wherein the windowing of samples of the incoming telephone signal is performed over 160 sample blocks.

5. The method according to claim 4 of windowing, wherein the windowing of samples of the incoming telephone signal is performed for a period of 20 milliseconds.

6. The method according to claim 1 for detecting the presence of ringer signals, further comprising the steps: normalizing the series of higher order coefficients with respect to the fundamental coefficient; setting the value of the generated fundamental coefficient to unity; and setting the value of each normalized coefficient to its absolute value.

7. The method according to claim 1 for detecting the presence of ringer signals, further comprising the step:

-8- deriving the value such that computed sums which exceed this threshold are identifiable as correlated ringer signals and computed sums which are less than this threshold are identifiable as uncorrelated non-ringer signals.

8. An apparatus for detecting the presence of ringer signals in an incoming telephone signal, the apparatus comprising: analysis means connected to the incoming telephone signal for receiving the telephone signal and generating a set of auto-correlation coefficients, including a fimdamental coefficient and a set of higher order coefficients; computing means connected to the analysis means for receiving and summing the coefficient values, wherein the sum of the coefficient values represents the correlation between frequency components of the incoming telephone signal; and comparing means connected to the computing means for receiving and comparing the sum produced in the computing means to a prestored threshold value and for producing an output signal indicating the relative correlation within the incoming telephone signals.

9. The apparatus for detecting the presence of ringer signals as in claim 8 further comprising: normalizing means connected between the analysis means and the computing means for receiving the auto-correlation coefficients from the analysis means and for normalizing the higher order coefficients with respect to the fundamental coefficient prior to transmitting the auto-correlation coefficients to the computing means.

10. The apparatus for detecting the presence of ringer signals as in claim 8, further comprising: absolute value means connected between the analysis means and the computing means for receiving the normalized auto-correlation coefficients from the analysis means and for converting each coefficient to its absolute value prior to transmitting the auto-correlation coefficients to the computing means.

-9-