METHODS AND APPARATUS FOR EMBEDDING WATERMARKS
RELATED APPLICATIONS
[0001] This application claims the benefit of the filing date of U.S. Provisional Application No. 60/478,626, filed June 13, 2003, and the benefit of U.S. Provisional Application No. 60/571,258, filed May 14, 2004, the disclosures of which are hereby incorporated herein by reference in their entireties.
TECHNICAL FIELD
[0002] The present disclosure relates generally to media measurements, and more particularly, to methods and apparatus for embedding watermarks in a compressed digital data stream.
BACKGROUND
[0003] In modern television or radio broadcast stations, compressed digital data streams are typically used to carry video and/or audio data for transmission. For example, the Advanced Television Systems Committee (ATSC) standard for digital television (DTV) broadcasts in the United States adopted Moving Picture Experts Group (MPEG) standards (e.g., MPEG-1, MPEG-2, MPEG-3, MPEG-4, etc.) for carrying video content and Digital Audio Compression standards (e.g., AC-3, which is also known as Dolby Digital®) for carrying audio content (i.e., ATSC Standard: Digital Audio Compression (AC-3), Revision A, August 2001). The AC-3 compression standard is based on a perceptual digital audio coding technique that reduces the amount of data needed to reproduce the original audio signal while minimizing perceptible distortion. In particular, the AC-3 compression standard recognizes that the human ear is unable to perceive
changes in spectral energy at particular spectral frequencies that are smaller than the masking energy at those spectral frequencies. The masking energy is a characteristic of an audio segment dependent on the tonality and noise-like characteristic of the audio segment. Different known psycho-acoustic models may be used to determine the masking energy at a particular spectral frequency. Further, the NC-3 compression standard provides a multi-channel digital audio format (e.g., 5.1 channels format) for digital television (DTV), high definition television (HDTV), digital versatile discs (DVDs), digital cable, and satellite transmissions that enables the broadcast of special sound effects (e.g., surround sound).
[0004] Existing television or radio broadcast stations employ watermarking techniques to embed watermarks within video and/or audio data streams compressed in accordance with compression standards such as the AC-3 compression standard and the MPEG Advanced Audio Coding (AAC) compression standard. Typically, watermarks are digital data that uniquely identify broadcasters and/or programs. Watermarks are typically extracted using a decoding operation at one or more reception sites (e.g., households or other media consumption sites) and, thus, may be used to assess the viewing behaviors of individual households and/or groups of households to produce ratings information.
[0005] However, many existing watermarking techniques are designed for use with analog broadcast systems. In particular, existing watermarking techniques convert analog program data to an uncompressed digital data stream, insert watermark data in the uncompressed digital data stream, and convert the watermarked data stream to an analog format prior to transmission. In the ongoing transition towards an all-digital broadcast environment in which compressed video and audio streams are transmitted by broadcast networks to local affiliates, watermark data may need to be embedded or inserted directly
in a compressed digital data stream. Existing watermarking techniques may decompress the compressed digital data stream into time-domain samples, insert the watermark data into the time-domain samples, and recompress the watermarked time-domain samples into a watermarked compressed digital data stream. Such decompression/compression may cause degradation in the quality of the media content in the compressed digital data stream. Further, existing decompression/compression techniques require additional equipment and cause delay of the audio component of a broadcast in a manner that, in some cases, may be unacceptable. Moreover, the methods employed by local broadcasting affiliates to receive compressed digital data streams from their parent networks and to insert local content through sophisticated splicing equipment prevent conversion of a compressed digital data stream to a time-domain (uncompressed) signal prior to recompression of the digital data streams.
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] FIG. 1 is a block diagram representation of an example media monitoring system.
[0007] FIG. 2 is a block diagram representation of an example watermark embedding system.
[0008] FIG. 3 is a block diagram representation of an example uncompressed digital data stream associated with the example watermark embedding system of FIG. 2.
[0009] FIG. 4 is a block diagram representation of an example embedding device that may be used to implement the example watermark embedding system of FIG. 2.
[0010] FIG. 5 depicts an example compressed digital data stream associated with the example embedding device of FIG. 4.
[0011] FIG. 6 depicts an example quantization look-up table that may be used to implement the example watermark embedding system of FIG. 2.
[0012] FIG. 7 depicts another example uncompressed digital data stream that may be compressed and then processed using the example watermark embedding system of FIG.
2.
[0013] FIG. 8 depicts an example compressed digital data stream associated with the example uncompressed digital data stream of FIG. 7.
[0014] FIG. 9 depicts one manner in which the example watermark embedding system of FIG. 2 may be configured to embed watermarks.
[0015] FIG. 10 depicts one manner in which the modification process of FIG. 9 may be implemented.
[0016] FIG. 11 depicts one manner in which a data frame may be processed.
[0017] FIG. 12 depicts one manner in which a watermark may be embedded in a compressed digital data stream.
[0018] FIG. 13 depicts an example code frequency index table that may be used to implement the example watermark embedding system of FIG. 2.
[0019] FIG. 14 is a block diagram representation of an example processor system that may be used to implement the example watermark embedding system of FIG. 2.
DETAILED DESCRIPTION
[0020] In general, methods and apparatus for embedding watermarks in compressed digital data streams are disclosed herein. The methods and apparatus disclosed herein may be used to embed watermarks in compressed digital data streams (i.e., without prior decompression of the compressed digital data streams). As a result, the methods and apparatus disclosed herein eliminate the need to subject compressed digital data streams
to multiple decompression/compression cycles, which are typically unacceptable to affiliates of, for example, television broadcast networks (i.e., multiple decompression/compression cycles may significantly degrade the quality of media content in the compressed digital data streams).
[0021] Prior to broadcast, for example, the methods and apparatus disclosed -herein may be used to unpack the modified discrete cosine transform (MDCT) coefficient sets associated with a compressed digital data stream formatted according to a digital audio compression standard such as the AC-3 compression standard. The mantissas of the unpacked MDCT coefficient sets may be modified to embed watermarks that imperceptibly augment the compressed digital data stream. Upon receipt of the compressed digital data stream, a receiving device (e.g., a set top television metering device at a media consumption site) may extract the embedded watermark inforroation from an uncompressed analog output such as, for example, output emanating from speakers of a television set. The extracted watermark information may be used to identify the media sources and/or programs (e.g., broadcast stations) associated with media currently being consumed (e.g., viewed, listened to, etc.) at a media consumption, site. In turn, the source and program identification information may be used in known manners to generate ratings information and/or any other information that may be used to assess the viewing behaviors associated with individual households and/or groups of households. [0022] Referring to FIG. 1 , an example broadcast system 100 including a service provider 110, a television 120, a remote control device 125, and a receiving device 130 is metered using an audience measurement system. The components of the broadcast system 100 may be coupled in any well-known manner. For example, the television 120 is positioned in a viewing area 150 located within a household occupied by one or more people, referred to as household members 160, some or all of whom have agreed to
participate in an audience measurement research study. The receiving device 130 may be a set top box (STB), a video cassette recorder, a digital video recorder, a personal video recorder, a personal computer, a digital video disc player, etc. coupled to the television 120. The viewing area 150 includes the area in which the television 120 is located and from which the television 120 may be viewed by the one or more household members 160 located in the viewing area 150.
[0023] In the illustrated example, a metering device 140 is configured to identify viewing information based on video/audio output signals conveyed from the receiving device 130 to the television 120. The metering device 140 provides this viewing information as well as other tuning and/or demographic data via a network 170 to a data collection facility 180. The network 170 may be implemented using any desired- combination of hardwired and wireless communication links, including for example, the Internet, an Ethernet connection, a digital subscriber line (DSL), a telephone line, a cellular telephone system, a coaxial cable, etc. The data collection facility 180 may be configured to process and/or store data received from the metering device 140 to produce ratings information.
[0024] The service provider 110 may be implemented by any service provider such as, for example, a cable television service provider 112, a radio frequency (RF) television service provider 114, and/or a satellite television service provider 116. The television 120 receives a plurality of television signals transmitted via a plurality of channels by the service provider 110 and may be adapted to process and display television signals provided in any format such as a National Television Standards Committee (NTSC) television signal format, a high definition television (HDTV) signal format, an Advanced Television Systems Committee (NTSC) television signal format, a phase alternation line (PAL) television signal format, a digital video broadcasting (DVB) television signal
format, an Association of Radio Industries and Businesses (ARIB) television signal format, etc.
[0025] The user-operated remote control device 125 allows a user (e.g., the household member 160) to cause the television 120 to tune to and receive signals transmitted on a desired channel, and to cause the television 120 to process and present or deliver the programming or media content contained in the signals transmitted on the desired channel. The processing performed by the television 120 may include, for example, extracting a video and/or an audio component delivered via the received signal, causing the video component to be displayed on a screen/display associated with the television 120, and causing the audio component to be emitted by speakers associated with the television 120. The programming content contained in the television signal may include, for example, a television program, a movie, an advertisement, a video game, a web page, a still image, and/or a preview of other programming content that is currently offered or will be offered in the future by the service provider 110.
[0026] While the components shown in FIG. 1 are depicted as separate structures within the broadcast system 100, the functions performed by some of these structures may be integrated within a single unit or may be implemented using two or more separate components. For example, although the television 120 and the receiving device 130 are depicted as separate structures, the television 120 and the receiving device 130 may be integrated into a single unit (e.g., an integrated digital television set). In another example, the television 120, the receiving device 130, and/or the metering device 140 may be integrated into a single unit.
[0027] To assess the viewing behaviors of individual household members 160 and/ox groups of households, a watermark embedding system (e.g., the watermark embedding system 200 of FIG. 2) may encode watermarks that uniquely identify broadcasters and/or
programs in the broadcast signals from the service providers 110. The watermark embedding system may be implemented at the service provider 110 so that each of the plurality of media signals (e.g., television signals) transmitted by the service provider 110 includes one or more watermarks. Based on selections by the household members 160, the receiving device 130 may tune to and receive media signals transmitted on a desired channel and cause the television 120 to process and present the programming content contained in the signals transmitted on the desired channel. The metering device 140 may identify watermark information based on video/audio output signals conveyed from the receiving device 130 to the television 120. Accordingly, the metering device 140 may provide this watermark information as well as other tuning and/or demographic data to the data collection facility 180 via the network 170.
[0028] In FIG. 2, an example watermark embedding system 200 includes an embedding device 210 and a watermark source 220. The embedding device 210 is configured to insert watermark information 230 from the watermark source 220 into a compressed digital data stream 240. The compressed digital data stream 240 may be compressed according to an audio compression standard such as the AC-3 compression standard and/or the MPEG-AAC compression standard, either of which may be used to process blocks of an audio signal using a predetermined number of digitized samples from each block. The source of the compressed digital data stream 240 (not shown) may be sampled at a rate of, for example, 48 kilohertz (kHz) to form audio blocks as described below.
[0029] Typically, audio compression techniques such as those based on the AC-3 compression standard use overlapped audio blocks and the MDCT algorithm to convert an audio signal into a compressed digital data stream (e.g., the compressed digital data stream 240 of FIG. 2). Two different block sizes (i.e., short and long blocks) may be used
depending on the dynamic characteristics of the sampled audio signal. For example, AC- 3 short blocks may be used to minimize pre-echo for transient segments of the audio signal and AC-3 long blocks may be used to achieve high compression gain for non- transient segments of the audio signal. In accordance with the AC-3 compression standard an AC-3 long block is derived from a block of 512 time-domain audio samples by means of an MDCT algorithm. In the case of the AC-3 long block, the 512 time- domain samples are obtained by concatenating a preceding (old) block of 256 time- domain samples and a current (new) block of 256 time-domain samples to create an audio block of 512 time-domain samples. The AC-3 long block contains 256 transform coefficients. In accordance with the same standard an AC-3 short block is derived from a pair of overlapping consecutive time-domain sample blocks of audio each containing 256 samples. The AC-3 short block contains the MDCT coefficients of these time-domain short blocks. Each time domain short block produces 128 coefficients resulting in a combined total of 256 coefficients. Thus, both the AC-3 long and AC-3 short blocks contain the same number of MDCT coefficients. In accordance with the MPEG-AAC compression standard as another example, a short block contains 128 samples and a long block contains 1024 samples. As is known, audio compression techniques decrease the number of bits required to represent an audio signal.
[0030] In the example of FIG. 3, an uncompressed digital data stream 300 includes a plurality of 256-sample time-domain audio blocks 310, generally shown as A0, Al, A2, A3, A4, and A5. The MDCT algorithm processes the audio blocks 310 to generate MDCT coefficient sets 320, generally shown as MAO, MAI, MA2, MA3, MA4, and MA5 (where MA5 is not shown). For example, the MDCT algorithm may process the audio blocks A0 and Al to generate the MDCT coefficient set MAO. The audio blocks A0 and Al are concatenated to generate a 512-sample audio block that is MDCT
transformed using the MDCT algorithm to generate the MDCT coefficient set MAO which includes 256 MDCT coefficients. Similarly, the audio blocks Al and A2 may be processed to generate the MDCT coefficient set MAI. Thus, the audio block Al is an overlapping audio block because it is used to generate both MDCT coefficient sets MAO and MAI . In a similar manner, the MDCT algorithm is used to transform the audio blocks A2 and A3 to generate the MDCT coefficient set MA2, the audio blocks A3 and A4 to generate the MDCT coefficient set MA3, the audio blocks A4 and A5 to generate the MDCT coefficient set MA4, etc. Thus, the audio block A2 is an overlapping audio block used to generate the MDCT coefficient sets MAI and MA2, the audio block A3 is an overlapping audio block used to generate the MDCT coefficient sets MA2 and MA3, the audio block A4 is an overlapping audio block used to generate the MDCT coefficient sets MA3 and MA4, etc. Together, the MDCT coefficient sets 320 form the compressed digital data stream 240.
[0031] As described in detail below, the embedding device 210 of FIG. 2 may embed or insert the watermark information or watermark 230 from the watermark source 220 into the compressed digital data stream 240. The watermark 230 may be used, for example, to uniquely identify broadcasters and/or programs so that media consumption information (e.g., viewing information) and/or ratings information may be produced. Accordingly, the embedding device 210 produces a watermarked compressed digital data stream 250 for transmission.
[0032] In the example of FIG. 4, the embedding device 210 includes an identifying unit 410, an unpacking unit 420, a modification unit 430, and a repacking unit 440. While the operation of the embedding device 210 is described below in accordance with the AC-3 compression standard, the embedding device 210 may be implemented to operate with additional or other compression standards such as, for example, the MPEG-
AAC compression standard. The operation of the embedding device 210 is described in greater detail in connection with FIG. 5.
[0033] To begin, the identifying unit 410 is configured to identify one or more frames 510 associated with the compressed digital data stream 240, a portion of which is shown by way of example as Frame A and Frame B in FIG. 5. As mentioned previously, the compressed digital data stream 240 may be a digital data stream compressed in accordance with the AC-3 standard (hereinafter "AC-3 data stream"). While the AC-3 data stream 240 may include multiple channels, for purposes of clarity, the following example describes the AC-3 data stream 240 as including only one channel. In the AC-3 data stream 240, each of the frames 510 includes a plurality of MDCT coefficient sets 520. In accordance with the AC-3 compression standard, for example, each of the frames 510 includes six MDCT coefficient sets (i.e., six "audblk"). For example, Frame A includes the MDCT coefficient sets MAO, MAI, MA2, MA3, MA4 and MA5 and Frame B includes the MDCT coefficient sets MB0, MB1, MB2, MB3, MB4 and MB5. [0034] The identifying unit 410 is also configured to identify header information associated with each of the frames 510, such as, for example, the number of channels associated with the AC-3 data stream 240. While the example AC-3 data stream 240 includes only one channel as noted above, an example compressed digital data stream having multiple channels is described below in connection with FIGS. 7 and 8. [0035] Returning to FIG. 5, the unpacking unit 420 is configured to unpack the MDCT coefficient sets 520 to determine compression information such as, for example, the parameters of the original compression process (i.e., the manner in which an audio compression technique compressed an audio signal or audio data to form the compressed digital data stream 240). For example, the unpacking unit 420 may determine how many bits are used to represent each of the MDCT coefficients within the MDCT coefficient
sets 520. Additionally, compression parameters may include information that limits the extent to which the AC-3 data stream 240 may be modified to ensure that the media content conveyed via the AC-3 data stream 240 is of a sufficiently high quality level. The embedding device 210 subsequently uses the compression information identified by the unpacking unit 420 to embed/insert the desired watermark information 230 into the AC-3 data stream 240 thereby ensuring that the watermark insertion is performed in a manner consistent with the compression information supplied in the signal. [0036] As described in detail in the AC-3 compression standard, the compression information also includes a mantissa and an exponent associated with each MDCT coefficient. Specifically, the presence of audio energy Et either at a particular frequency k (e.g., a tone) or spread across a band of frequencies proximate to the particular frequency k (e.g., a noise-like characteristic) creates a masking effect. That is, the human ear is unable to perceive a change in energy in a spectral region either at a frequency k or spread across the band of frequencies proximate to the frequency A: if that change is less than a given energy threshold ΔE/t. Because of this characteristic of the human ear, an MDCT coefficient m# associated with the frequency k may be quantized with a step size related to ΔEt without risk of causing any humanly perceptible changes to the audio content. For the AC-3 data stream 240, each MDCT coefficient is unpacked as a mantissa Mk and an exponent Xk such that m^ = Mk .2 ' t- The number of bits used to represent the mantissa Mk of each MDCT coefficient of the MDCT coefficient sets 520 may be determined based on known quantization look-up tables published in the AC-3 compression standard (e.g., the quantization look-up table 600 of FIG. 6). In the example of FIG. 6, the quantization look-up table 600 provides mantissa codes or bit patterns and mantissa values for MDCT coefficients represented by a four-bit number. As described in detail below, the mantissa Mk may be changed (e.g., augmented) to represent a
modified value of an MDCT coefficient to embed a watermark in the AC-3 data stream 240.
[0037] Returning to FIG. 5, the modification unit 430 is configured to perform an inverse transform of each of the MDCT coefficient sets 520 to generate time-domain audio blocks 530, shown by way of example as TAO', TA3", TA4', TA4", TA5', TA5", TBO', TBO", TBl ', TBl", and TB5' (TAO" through TA3' and TB2' through TB4" are not shown). The modification unit 430 performs inverse transform operations to generate sets of previous (old) time-domain audio blocks (which are represented as prime blocks) and sets of current (new) time-domain audio blocks (which are represented as double- prime blocks) associated with the 256-sample time-domain audio blocks that were concatenated to form the MDCT coefficient sets 520 of the AC-3 data stream 240. For example, the modification unit 430 performs an inverse transform on the MDCT coefficient set MA5 to generate time-domain blocks TA4" and TA5', the MDCT coefficient set MB0 to generate TA5" and TBO', the MDCT coefficient set MB1 to generate TBO" and TBl ', etc. In this manner, the modification unit 430 generates reconstructed time-domain audio blocks 540, which provide a reconstruction of the original time-domain audio blocks that were compressed to form the AC-3 data stream 240. To generate the reconstructed time-domain audio blocks 540, the modification unit 430 may add time-domain audio blocks based on, for example, the known Princen- Bradley time domain alias cancellation (TDAC) technique as described in Princen et al., Analysis/Synthesis Filter Bank Design Based on Time Domain Aliasing Cancellation, Institute of Electrical and Electronics Engineers (IEEE) Transactions on Acoustics, Speech and Signal Processing, Vol. ASSP-35, No. 5, pp. 1153 - 1161 (1996). For example, the modification unit 430 may reconstruct the time-domain audio block TA5 (i.e., TA5R) by adding the prime time-domain audio block TA5' and the double-prime
time-domain audio block TA5" using the Princen-Bradley TDAC technique. Likewise, the modification unit 430 may reconstruct the time-domain audio block TBO (i.e., TB0R) by adding the prime audio block TBO' and the double-prime audio block TBO" using the Princen-Bradley TDAC technique. In this manner, the original time-domain audio blocks used to form the AC-3 data stream 240 are reconstructed (i.e., the reconstructed time- domain audio blocks 540) without having to decompress the AC-3 data stream 240 which, as described in detail below, enables the watermark 230 to be embedded or inserted directly into the AC-3 data stream 240.
[0038] The modification unit 430 is also configured to insert the watermark 230 into the reconstructed time-domain audio blocks 540 to generate watermarked time-domain audio blocks 550, generally shown as TA0W, TA4W, TA5W, TB0W, TB1W and TB5W (blocks TA1W, TA2W, TA3W, TB2W, TB3W and TB4W are not shown). To insert the watermark 230, the modification unit 430 generates a modifiable time-domain audio block by concatenating two adjacent reconstructed time-domain audio blocks to create a 512-sample audio block. For example, the modification unit 430 may concatenate the reconstructed time-domain audio blocks TA5R and TB0R (each being a 256-sample audio block) to form a 512-sample audio block. The modification unit 430 may then insert the watermark 230 into the 512-sample audio block formed by the reconstructed time-domain audio blocks TA5R and TB0R to generate the watermarked time-domain audio blocks TA5W and TB0W. Encoding processes such as those described in U.S. Patent Nos. 6,272,176, 6,504,870, and 6,621,881 may be used to insert the watermark 230 into the reconstructed time-domain audio blocks 540. The disclosures of U.S. Patent Nos. 6,272,176, 6,504,870, and 6,621,881 are hereby incorporated by reference herein in their entireties.
[0039] In the example encoding methods and apparatus described in U.S. Patent Nos. 6,272,176, 6,504,870, and 6,621,881, watermarks may be inserted into a 512-sample audio block. For example, each 512-sample audio block carries one bit of embedded or inserted data of the watermark 230. In particular, spectral frequency components with indices fi and f
2 may be modified or augmented to insert data bits associated with the watermark 230. To insert a binary "1," for example, a power at the first spectral frequency associated with the index fi may be increased or augmented to be a spectral power maximum within a frequency neighborhood (e.g., a frequency neighborhood- defined by the indices fi - 2, fi — 1, f
l9 fi + 1, and
+ 2). At the same time, the power at the second spectral frequency associated with the index f
2 is attenuated or augmented to be a spectral power minimum within a frequency neighborhood (e.g., a frequency neighborhood defined by the indices f
2 - 2, f
2 - 1, f
2, f
2 + 1, and f
2 + 2). Conversely, to insert a binary "0," the power at the first spectral frequency associated with the index fi is attenuated to be a local spectral power minimum while the power at the second spectral frequency associated with the index f
2 is increased to a local spectral power maximum. [0040] Returning to FIG. 5, based on the watermarked time-domain audio blocks 550, the modification unit 430 generates watermarked MDCT coefficient sets 560, shown by way of example as MA0W, MA4W, MA5W, MB0W and MB5W (blocks MA1W, MA2W, MA3W, MB1W, MB2W, MB3W and MB4W are not shown). Following the example described above, the modification unit 430 generates the watermarked MDCT coefficient set MA5W based on the watermarked time-domain audio blocks TA5 W and TB0W. Specifically, the modification unit 430 concatenates the watermarked time- domain audio blocks TA5W and TB0W to form a 512-sample audio block and converts the 512-sample audio block into the watermarked MDCT coefficient set MA5W which,
as described in greater detail below, may be substituted for the original MDCT coefficient set MA5.
[0041] The difference between the MDCT coefficient sets 520 and the watermarked MDCT coefficient sets 560 represents a change in the AC-3 data stream 240 as a result of embedding or inserting the watermark 230. As described in connection with FIG. 6, for example, the modification unit 430 may replace the MDCT coefficient set MA5 with its corresponding watermarked MDCT coefficient set MA5W. Quantization look-up tables (e.g., the look-up table 600 of FIG. 6) may be used to determine new mantissa values associated with the MDCT coefficients of the watermarked MDCT coefficient sets 560 to replace the old mantissa values associated with the MDCT coefficients of the MDCT coefficient sets 520. Thus, the new mantissa values represent the change in or augmentation of the AC-3 data stream 240 as a result of embedding or inserting the watermark 230.
[0042] Turning to FIG. 6, the example quantization look-up table 600 includes mantissa codes and mantissa values for a fifteen-level quantization of an example mantissa i in the range of -0.9333 to +0.9333. While the example quantization look-up table 600 provides mantissa information associated with MDCT coefficients that are represented using four bits, the AC-3 compression standard provides quantization look-up tables associated with other suitable numbers of bits per MDCT coefficient. To illustrate one manner in which the modification unit 430 may modify a particular MDCT coefficient m with a mantissa Mk contained in the MDCT coefficient set MA5, assume the original mantissa value is -0.2666 (i.e., -4/15). Using the quantization look-up table 600, the mantissa code corresponding to the particular MDCT coefficient mk in the MDCT coefficient set MA5 is determined to be 0101. The watermarked MDCT coefficient set MA5W includes a watermarked MDCT coefficient wmt with a mantissa
value WMk.. Further, assume the new mantissa value of the corresponding watermarked MDCT coefficient wm of the watermarked MDCT coefficient set MA5W is -0.4300, which lies between the mantissa codes of 0011 and 0100. In other words, the watermark 230, in this example, results in a difference of -0.1667 between the original mantissa value of -0.2666 and the watermarked mantissa value of -0.4300. [0043] To embed or insert the watermark 230 in the AC-3 data stream 240, the modification unit 430 may substitute the watermarked MDCT coefficient set MA5W for the MDCT coefficient set MA5 by modifying or augmenting the MDCT coefficients in the MDCT coefficient set MA5. Continuing with above example, either mantissa code 0011 or mantissa code 0100 may replace the mantissa code 0101 associated with the MDCT coefficient ntk (i.e., an MDCT coefficient in the MDCT coefficient set MA5) because the watermarked mantissa WMk associated with the watermarked MDCT coefficient wmk (i.e., an MDCT coefficient in the watermarked MDCT coefficient set MA5W) lies between the mantissa codes of 0011 and 0100 (i.e., the mantissa value corresponding to the watermarked MDCT coefficient wnik is -0.4300). The mantissa value corresponding to the mantissa code 0011 is -0.5333 (i.e., -8/15) and the mantissa value corresponding to the mantissa code 0100 is -0.4 (i.e., -6/15). In this example, the modification unit 430 selects the mantissa code 0100 instead of the mantissa code 0011 to replace the original mantissa code 0101 associated with the MDCT coefficient m£ because the mantissa value -0.4 corresponding to the mantissa code 0100 is closest to the desired watermark mantissa value -0.4300. As a result, the new mantissa bit pattern of 0100, which corresponds to the watermarked mantissa WMk of the watermarked MDCT coefficient wmk, replaces the original mantissa bit pattern of 0101. Likewise, each of the MDCT coefficients in the MDCT coefficient set MA5 may be modified in the manner described above. If a new (i.e., watermarked) mantissa value is outside the quantization
range of mantissa values (i.e., greater than 0.9333 or less than -0.9333), either the positive limit of 1110 or the negative limit of 0000 is selected as the new mantissa code. Additionally, while the mantissa codes associated with each MDCT coefficient of an MDCT coefficient set may be modified as described above, the exponents associated with the MDCT coefficients remain unchanged.
[0044] The repacking unit 440 is configured to repack the watermarked MDCT coefficient sets 560 associated with each frame of the AC-3 data stream. 240 for transmission. In particular, the repacking unit 440 identifies the position of each MDCT coefficient set within a frame of the AC-3 data stream 240 so that the corresponding watermarked MDCT coefficient set can be used to replace the MDCT coefficient set. To rebuild a watermarked version of Frame A, for example, the repacking "unit 440 may identify the position of the MDCT coefficient sets MAO to MA5 and substitute the corresponding watermarked MDCT coefficient sets MA0W to MA5W in the corresponding identified positions. Using the unpacking, modifying, and repacking processes described herein, the AC-3 data stream 240 remains a compressed digital data stream while the watermark 230 is embedded or inserted in the AC-3 data stream 240. As a result, the embedding device 210 inserts the watermark 230 into the A.C-3 data stream 240 without additional decompression/compression cycles that may degrade the quality of the media content in the AC-3 data stream 240.
[0045] For simplicity, the AC-3 data stream 240 is described in connection with FIG. 5 to include a single channel. However, the methods and apparatus disclosed herein may be applied to compressed digital data streams having audio blocks associated with multiple channels, such as 5.1 channels (i.e., five full-bandwidth channels), as described below. In the example of FIG. 7, an uncompressed digital data stream 700 may include a plurality of audio block sets 710. Each of the audio block sets 710 may include audio
blocks associated with multiple channels 720 and 730 including, for example, a front left channel, a front right channel, a center channel, a surround left channel, a surround right channel, and a low-frequency effect (LFE) channel (e.g., a sub-woofer channel). For example, the audio block set AUD0 includes an audio block AOL associated with the front left channel, an audio block A0R associated with the front right channel, an audio block AOC associated with the center channel, an audio block A0SL associated with the surround left channel, an audio block A0SR associated with the surround right channel, and an audio block A0LFE associated with the LFE channel. Similarly, the audio block set AUDI includes an audio block AIL associated with the front left channel, an audio block AIR associated with the front right channel, an audio block A1C associated with the center channel, an audio block A1SL associated with the surround left channel, an audio block AISR associated with the surround right channel, and an audio block AILFE associated with the LFE channel.
[0046] Each of the audio blocks associated with a particular channel in the audio block sets 710 may be processed in a manner similar to that described above in connection with FIGS. 5 and 6. For example, the audio blocks associated with the center channel 810 of FIG. 8, generally shown as AOC, A1C, A2C, and A3C, maybe transformed to generate the MDCT coefficient sets 820 associated with a compressed digital data stream 800. As noted above, each of the MDCT coefficient sets 820 may be derived from a 512-sample audio block formed by concatenating a preceding (old) 256- sample audio block and a current (new) 256-sample audio block. The MDCT algorithm may then process the time-domain audio blocks 810 (e.g., AOC through A5C) to generate the MDCT coefficient sets (e.g., M0C through M5C).
[0047] Based on the MDCT coefficient sets 820 of the compressed digital data stream 800, the identifying unit 410 identifies a plurality of frames (not shown) and header
information associated with each of the frames as described above. The header information includes compression information associated with the compressed digital data stream 800. For each of the frames, the unpacking unit 420 unpacks the MDCT coefficient sets 820 to determine the compression information associated with the MDCT coefficient sets 820. For example, the unpacking unit 420 may identify the number of bits used by the original compression process to represent the mantissa of each MDCT coefficient in each of the MDCT coefficient sets 820. Such compression information may be used to embed the watermark 230 as described above in connection with FIG. 6. The modification unit 430 then generates inverse transformed time-domain audio blocks 830, generally shown as TA0C", TA1C, TA1C", TA2C, TA2C", and TA3C. The time- domain audio blocks 830 include a set of previous (old) time-domain audio blocks (which are represented as prime blocks) and a set of current (new) time-domain audio blocks (which are represented as double-prime blocks). By adding the corresponding prime blocks and double-prime blocks based on, for example, the Princen-Bradley TDAC technique, original time-domain audio blocks compressed to form the AC-3 digital data stream 800 may be reconstructed (i.e., the reconstructed time-domain audio blocks 840). For example, the modification unit 430 may add the time-domain audio blocks TA1C and TA1C" to reconstruct the time-domain audio block TA1C (i.e., TA1CR). Likewise, the modification unit 430 may add the time-domain audio blocks TA2C and TA2C" to reconstruct the time-domain audio block TA2C (i.e., TA2CR).
[0048] To insert the watermark 230 from the watermark source 220, the modification unit 430 concatenates two adjacent reconstructed time-domain audio blocks to create a 512-sample audio block (i.e., a modifiable time-domain audio block). For example, the modification unit 430 may concatenate the reconstructed time-domain audio blocks TA1CR and TA2CR, each of which is a 256-sample short block, to form a 512-sample
audio block. The modification unit 430 then inserts the watermark 230 into the 512- sample audio block formed by the reconstructed time-domain audio blocks TA1CR and TA2CR to generate the watermarked time-domain audio blocks TA1CW and TA2CW. [0049] Based on the watermarked time-domain audio blocks 850, the modification unit 430 may generate the watermarked MDCT coefficient sets 860. For example, the modification unit 430 may concatenate the watermarked time-domain audio blocks TA1CW and TA2CW to generate the watermarked MDCT coefficient set M1CW. The modification unit 430 modifies the MDCT coefficient sets 820 by replacing each with a corresponding one of the watermarked MDCT coefficient sets 860. For example, the modification unit 430 may substitute the watermarked MDCT coefficient set M1CW for the original MDCT coefficient set MIC. The modification unit 430 may then repeat the process described above for the audio blocks associated with each channel to insert the watermark 230 into the compressed digital data stream 800. [0050] FIG. 9 is a flow diagram depicting one manner in which the example watermark embedding system of FIG. 2 may be configured to embed or insert watermarks in a compressed digital data stream. The example process of FIG. 9 may be implemented as machine accessible instructions utilizing any of many different programming codes stored on any combination of machine-accessible media such as a volatile or nonvolatile memory or other mass storage device (e.g., a floppy disk, a CD, and a DVD). For example, the machine accessible instructions may be embodied in a machine-accessible medium such as a programmable gate array, an application specific integrated circuit (ASIC), an erasable programmable read only memory (EPROM), a read only memory (ROM), a random access memory (RAM), a magnetic media, an optical media, and/or any other suitable type of medium. Further, although a particular order of actions is illustrated in FIG. 9, these actions can be performed in other temporal sequences. Again,
the flow diagram 900 is merely provided and described in connection with the components of FIGS. 2 to 5 as an example of one way to configure a system to embed watermarks in a compressed digital data stream.
[0051] In the example of FIG. 9, the process begins with the identifying unit 410 (FIG. 4) identifying a frame associated with the compressed digital data stream 240 (FIG. 2) such as Frame A (FIG. 5) (block 910). The identified frame may include a plurality of MDCT coefficient sets formed by overlapping and concatenating a plurality of audio blocks. In accordance with the AC-3 compression standard, for example, a frame may include six MDCT coefficient sets (i.e., six "audblk"). Further, the identifying unit 410 (FIG. 4) also identifies header information associated with the frame (block 920). For example, the identifying unit 410 may identify the number of channels associated with the compressed digital data stream 240.
[0052] The unpacking unit 420 then unpacks the plurality of MDCT coefficient sets to determine compression information associated with the original compression process used to generate the compressed digital data stream 240 (block 930). In particular, the unpacking unit 420 identifies the mantissa Mk and the exponent^ of each MDCT coefficient nik of each of the MDCT coefficient sets. The exponents of the MDCT coefficients may then be grouped in a manner compliant with the AC-3 compression standard. The unpacking unit 420 (FIG. 4) also determines the number of bits used to represent the mantissa of each of the MDCT coefficients so that a suitable quantization look-up table specified by the AC-3 compression standard may be used to modify or augment the plurality of MDCT coefficient sets as described above in connection with FIG. 6. Control then proceeds to block 940 which is described in greater detail below in connection with FIG. 10.
[0053] As illustrated in FIG. 10, the modification process 940 begins by using the modifying unit 430 (FIG. 4) to perform an inverse transform of the MDCT coefficient sets to generate inverse transformed time-domain audio blocks (block 1010). In particular, the modification unit 430 generates a previous (old) time-domain audio block (which, for example, is represented as a prime block in FIG. 5) and a current (new) time- domain audio block (which is represented as a double-prime block in FIG. 5) associated with each of the 256-sample original time-domain audio blocks used to generate the corresponding MDCT coefficient set. As described in connection with FIG. 5, for example, the modification unit 430 may generate TA4" and TA5' from the MDCT coefficient set MA5, TA5" and TBO' from the MDCT coefficient set MBO, and TBO" and TBl ' from the MDCT coefficient set MB1. For each time-domain audio block, the modification unit 430 adds corresponding prime and double-prime blocks to reconstruct the time-domain audio block based on, for example, the Princen-Bradley TDAC technique (block 1020). Following the above example, the prime block TA5' and the double-prime block TA5" may be added to reconstruct the time-domain audio block TA5 (i.e., the reconstructed time-domain audio block TA5R) while the prime block TBO' and the double-prime block TBO" may be added to reconstruct the time-domain audio block TBO (i.e., the reconstructed time-domain audio block TB0R).
[0054] To insert the watermark 230, the modification unit 430 generates modifiable time-domain audio blocks using the reconstructed time-domain audio blocks (block 1030). The modification unit 430 generates a modifiable 512-sample time-domain audio block using two adjacent reconstructed time-domain audio blocks. For example, the modification unit 430 may generate a modifiable time-domain audio block by concatenating the reconstructed time-domain audio blocks TA5R and TB0R of FIG. 5.
[0055] Implementing an encoding process such as, for example, one or more of the encoding methods and apparatus described in U.S. Patent Nos. 6,272,176, 6,504,870, and/or 6,621,881, the modification unit 430 inserts the watermark 230 from the watermark source 220 into the modifiable time-domain audio blocks (block 1040). For example, the modification unit 430 may insert trie watermark 230 into the 512-sample time-domain audio block generated using the reconstructed time-domain audio blocks TA5R and TB0R to generate the watermarked time-domain audio blocks TA5W and TBOW. Based on the watermarked time-domain audio blocks and the compression information, the modification unit 430 generates watermarked MDCT coefficient sets (block 1050). As noted above, two watermarked time-domain audio blocks, where each block includes 256 samples, may be used to generate a watermarked MDCT coefficient set. For example, the watermarked time-domain audio blocks TA5W and TBOW may be concatenated and then used to generate the watermarked MDCT coefficient set MA5W. [0056] Based on the compression information associated with the compressed digital data stream 240, the modification unit 430 calculates the mantissa value associated with each of the watermarked MDCT coefficients in the watermarked MDCT coefficient set MA5W as described above in connection with FIG. 6. In this manner, the modification unit 430 can modify or augment the original MDCT coefficient sets using the watermarked MDCT coefficient sets to embed or insert the watermark 230 in the compressed digital data stream 240 (block 106O). Following the above example, the modification unit 430 may replace the original MDCT coefficient set MA5 with the watermarked MDCT coefficient set MA5W of FIG, 5. The modification unit 430 may replace an original MDCT coefficient in the MDCT coefficient set MA5 with a corresponding watermarked MDCT coefficient (which has an augmented mantissa value) from the watermarked MDCT coefficient set MA5 W. Alternatively, the modification
unit 430 may compute the difference between the mantissa codes associated with the original MDCT coefficient and the corresponding watermarked MDCT coefficient (i.e., AMk = Mk - WMk) and modify the original MDCT coefficient based on the difference M . In any case, after modifying the original MDCT coefficient sets, the modification process 940 terminates and returns control to block 950.
[0057] Referring back to FIG. 9, the repacking unit 440 repacks the frame of the compressed digital data stream (block 950). The repacking unit 440 identifies the position of the MDCT coefficient sets within the frame so that the watermarked MDCT coefficient sets may be substituted in the positions of the original MDCT coefficient sets to rebuild the frame. At block 960, if the embedding device 210 determines that additional frames of the compressed digital data stream 240 need to be processed, then control returns to block 910. If, instead, all frames of the compressed digital data stream 240 have been processed, then the process 900 terminates.
[0058] As noted above, known watermarking techniques typically decompress a compressed digital data stream into uncompressed time-domain samples, insert the watermark into the time-domain samples, and recompress the watermarked time-domain samples into a watermarked compressed digital data stream. In contrast, the digital data stream 240 remains compressed during the example unpacking, modifying, and repacking processes described herein. As a result, the watermark 230 is embedded into the compressed digital data stream 240 without additional decompression/compression cycles that may degrade the quality of the content in the compressed digital data stream 500. [0059] To further illustrate the example modification process 940 of FIGS. 9 and 10, FIG. 11 depicts one manner in which a data frame (e.g., an AC-3 frame) may be processed. The example frame processing process 1100 begins with the embedding device 210 reading the header information of the acquired frame (e.g., an AC-3 frame)
(block 1110) and initializing an MDCT coefficient set count to zero (block 1120). In the case where an AC-3 frame is being processed, each AC-3 frame includes six MDCT coefficient sets having compressed-domain data (e.g., MAO, MAI, MA2, MA3, MA4 and MA5 of FIG. 5, which are also known as "audblks" in the AC-3 standard). Accordingly, the embedding device 210 determines whether the MDCT coefficient set count is equal to six (block 1130). If the MDCT coefficient set count is not yet equal to six, thereby indicating that at least one more MDCT coefficient set requires processing the embedding device 210 extracts the exponent (block 1140) and the mantissa (block 1150) associated with an MDCT coefficient of the frame (e.g., the original mantissa Mk described above in connection with FIG. 6). The embedding device 210 computes a new mantissa associated with a code symbol read at block 1220 (e.g., the new mantissa WMk described above in connection with FIG. 6) (block 1160) and modifies the original mantissa associated with the frame based on the new mantissa (block 1170). For example, the original mantissa may be replaced with the new mantissa. The embedding device 210 increments the MDCT coefficient set count by one (block 1180) and control returns to block 1130. Although the example process of FIG. 11 is described above to include six MDCT coefficient sets (e.g., the threshold of the MDCT coefficient set count is six), a process utilizing more or fewer MDCT coefficient sets could be used instead. At block 1130, if the MDCT coefficient set count is equal to six, then all MDCT coefficient sets have been processed such that the watermark has been embedded and the embedding device 210 repacks the frame (block 1190).
[0060] As noted above, many methods are known to embed a watermark imperceptible to the human ear (e.g., an inaudible code) in an uncompressed audio signal. For example, one known method is described in U.S. Patent No. 6,421,445 to Jensen et al., the disclosure of which is hereby incorporated by reference herein in its entirety. In
particular, as described by Jensen et al., a code signal (e.g., a watermark) may include information at a combination often different frequencies, which are detectable by a decoder using a Fourier spectral analysis of a sequence of audio samples (e.g., a sequence of 12,288 audio samples as described in detail below). For example, an audio signal may be sampled at a rate of 48 kilo-Hertz (kHz) to output an audio sequence of 12,288 audio samples that may be processed (e.g., using a Fourier transform) to acquire a relatively high-resolution (e.g., 3.9 Hz) frequency domain representation of the uncompressed audio signal. However, in accordance with the encoding process of the method disclosed by Jensen et al., a sinusoidal code signal having constant amplitude across an entire sequence of audio samples is unacceptable because the sinusoidal code signal may be perceptible to the human ear. To satisfy the masking energy constraints (i.e., to ensure that the sinusoidal code signal information remains imperceptible), the sinusoidal code signal is synthesized across the entire sequence of 12,288 audio samples using a masking energy analysis which determines a local sinusoidal amplitude within each block of audio samples (e.g., wherein each block of audio samples may include 512 audio samples). Thus, the local sinusoidal waveforms may be coherent (in-phase) across the sequence of 12,288 audio samples.
[0061] In contrast to the method disclosed by Jensen et al., the methods and apparatus described herein may be used to embed a watermark or other code signal in a compressed audio signal in a manner such that a compressed digital data stream containing the compressed audio signal remains compressed during the unpacking, modifying, and repacking processes. FIG. 12 depicts one manner in which a watermark, such as that disclosed by Jensen et al., may be inserted in a compressed audio signal. The example process 1200 begins with initializing a frame count to zero (block 1210). Eight frames (e.g., AC-3 frames) representing a total of 12,288 audio samples of each audio channel
may be processed to embed one or more code symbols (e.g., one or more of the symbols "0", "1", "S", and "E" shown in FIG. 13) into the audio signal. Although the compressed digital data stream is described herein to include 12,288 audio samples, the compressed digital data stream may have more or less audio samples. In particular, the embedding device 210 (FIG. 2) may read a watermark 230 from the watermark source 220 to inject one or more code symbols into the sequence of frames (block 1220). The embedding device 210 may acquire one of the frames (block 1230) and proceed to the frame processing operation 1100 described above to process the acquired frame. Accordingly, the example frame processing operation 1100 terminates and control returns to block 1250 to increment the frame count by one. The embedding device 210 determines whether the frame count is eight (block 1260). If the frame count is not eight, the embedding device 210 returns to acquire another frame in the sequence and repeat the example frame processing operation 1100 as described above in connection with FIG. 11 to process another frame. If, instead, the frame count is eight, the embedding device 210 returns to block 1210 to reinitialize the frame count to zero and repeat the process 1200 to process another sequence of frames.
[0062] As noted above, a code signal (e.g., the watermark 230) may be embedded or injected into the compressed digital data stream (e.g., an AC-3 data stream). As shown in the example table 1300 of FIG. 13, the code signal may include a combination often sinusoidal components corresponding to
to represent one of four code symbols, generally shown as "0," "1," "S," and "E." In particular, the code symbol "0" may represent a binary value of zero and the code symbol "1" may represent a binary value of one. Further, the code symbol "S" may represent the start of a message and the code symbol "E" may represent the end of a message. While only four code symbols are shown in FIG. 13, more or fewer code symbols could be used instead.
Further, the 512-sample central frequency indexes (e.g., 10, 12, 14, 16, 18, 20, 22, 24-, 26, and 28) are associated with a low resolution frequency domain representation of the compressed digital data stream and the 12,288-sample central frequency indexes (e.g.,
240, 288, 336, 384, 432, 480, 528, 576, 624, and 672) are associated with a high resolution frequency domain representation of the compressed digital data stream. [0063] As noted above, each code symbol may be formed using ten sinusoidal components associated with the frequency indexes f\ through/! o depicted in table 13O0.
As shown in the example table 1300, each of the frequency indexes f\ through/Jo has a unique frequency value at or proximate to each of the 12,288-sample central frequency indexes. For example, a code signal for injecting or embedding the code symbol "0" includes ten sinusoidal components corresponding to the frequency indexes 237, 289-,
339, 383, 429, 481, 531, 575, 621, and 673, respectively. Likewise, a code signal for injecting or embedding the code symbol "1" includes ten sinusoidal components corresponding to the frequency indexes 239, 291, 337, 381, 431, 483, 529, 573, 623, and
675, respectively.
[0064] Each of the ten sinusoidal components associated with the frequency indexes f
\ through
ϊo may be synthesized in the time domain using the methods and apparatus described herein. For example, the code signal for injecting or embedding the code symbol "0" may include sinusoids cι(k), c
2(&), c
3(A), c (&), c
5(&), c
6(A), cη(k), c$(k), cgCJ), and c
\o(k). In particular, the first sinusoid c
\(k) may be synthesized in the time domain as 2π*2 1k a sequence of samples as follows: c
x (k) = cos for k = 0 through 12287. In 12288 addition, the sample values in a 512-sample audio block (e.g., a long block) associated with the first sinusoid
may be computed as follows:
2π * 237 * ( p * 256 + m) c, (m) = w(m) cos ; — — for m = 0 through 511 and/? = 0 through -46, 12288
where w(m) is the window function used in the AC-3 compression described above. The
MDCT transform of clp (m) may include a number of MDCT coefficient values (e.g., 256
real numbers). In this example, the MDCT coefficient values associated with the 512- sample frequency indexes 9, 10, and 11 may have significant magnitudes because c\p(m) is associated with the 12,288-sample central frequency index 240, which corresponds to the 512-sample central frequency index 10. The MDCT coefficient values associated with other 512-sample frequency indexes are negligible relative to the MDCT coefficient values associated with the 512-sample frequency indexes 9, 10, and 11. The MDCT coefficient values associated with c\p(m) may be divided by a normalization factor Q as
512 follows: Q = = 128 , where 512 is a number of samples associated with each block.
As a result, a time-domain cosine wave of unit amplitude at the 12,288-sample central frequency index 240 may produce a unit amplitude MDCT coefficient at the 512-sample central frequency index 10 to serve as a reference for the MDCT coefficients associated with the 512-sample audio block having a central frequency of approximately 936 Hz (as shown in FIG. 13).
[0065] For cλp (m) , the code frequency index 237 (e.g., the frequency value
corresponding to the frequency index/] associated with the code symbol "0") causes the 512-sample central frequency index 10 to have the highest MDCT magnitude relative to the 512-sample frequency indexes 9 and 11 because the 512-sample central frequency index 10 corresponds to the 12,288-sample central frequency index 240 and the code frequency index 237 is proximate to the 12,288-sample central frequency index 240. Likewise, the second frequency index fi corresponding to the code frequency index 289 may produce MDCT coefficients with significant MDCT magnitudes in the 512-sample frequency indexes 11, 12, and 13. The code frequency index 289 may cause the 512-
sample central frequency index 12 to have the highest MDCT magnitude because the 512-sample central frequency index 12 corresponds to the 12,288-sample central frequency index 288 and the code frequency index 289 is proximate to the 12,288-sample central frequency index 288. In yet another example, the third frequency index/ corresponding to the code frequency index 339 may produce MDCT coefficients with significant MDCT magnitudes in the 512-sample frequency indexes 13, 14, and 15. The code frequency index 339 may cause the 512-sample central frequency index 14 to have the highest MDCT magnitude because the 512-sample central frequency index 14 corresponds to the 12,288-sample central frequency index 336 and the code frequency index 339 is proximate to the 12,288-sample central frequency index 336. Based on the unit amplitude cosine waves of each of the ten frequency indexes/ through/o associated with the code symbol "0," the MDCT coefficient representing the actual code signal may be computed for 512-sample frequency indexes ranging from 9 to 29. Some of the 512- sample frequency indexes such as, for example, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, and 29 may be influenced by two neighboring code frequency indexes because none of these 512-sample frequency indexes corresponds directly to a 12,288-sample central frequency index (e.g., 240, 288, 336, 384, 432, 480, 528, 576, 624, or 672). Accordingly, in each 512-sample audio block of the compressed digital data stream, the MDCT coefficient may be computed as described below to represent the code signal. [0066] In the compressed AC-3 data stream, for example, each AC-3 frame includes MDCT coefficient sets having six MDCT coefficients (e.g., MAO, MAI, MA2, MA3, MA4, and MA5 of FIG. 5) with each MDCT coefficient corresponding to a 512-sample audio block. As described above in connection with FIGS. 5 and 6, each MDCT
coefficient is represented as mk = M k * 2~x" = ($k*Nk ) * 2~Xk , where Xk is the exponent
and Mk is the mantissa. The mantissa Mk is a product of a mantissa step size sk and an
integer value Nk . The mantissa step size sk and the exponent Xk may be used to form a
quantization step sizeSA = sk * 2~Xl . Referring to the look-up table 600 of FIG. 6, for
example, the mantissa step size sk is 1/15 and the integer value Nk is -4 when the
original mantissa value is -0.2666 (i.e., -4/15).
[0067] To inject a code signal into the compressed AC-3 data stream, modifications to the mantissa M
k for k - 9 through 29 are determined. For example, assume the MDCT coefficient magnitudes C
9,
corresponding to the MDCT coefficients m , mw,
and mu are -0.3, 0.8, and 0.2, respectively. The code MDCT magnitude C associated
with the 512-sample central frequency index 11 is the MDCT coefficient having the lowest magnitude (e.g., an absolute value of 0.2). The value of the code MDCT
magnitude Cn is used to obtain the values of the MDCT coefficients m mio, and mπ
(e.g., -1.5, 4.0, and 1.0, respectively) because the code MDCT magnitude Cn has the
lowest magnitude. In particular, the mantissa integer value Nn is increased by 1 to
normalize the values of the MDCT coefficients m m o, and mπ. As a result, the mantissa integer values N9 and o are modified as follows because the step size for the
— 1 5 *S 4 0* S mantissa Mn is Sn : N9- > N9 + — : — and N10 - > N10 + — — . Thus, the S9 S10
modified mantissa integer values N , Nio, and Nπ may be used to generate corresponding modified MDCT coefficients wm wm o, and wm\\, respectively. As described in connection with FIGS. 5, and 6, the modified MDCT coefficients wm wmio, and wm\ \ may be substituted for the original MDCT coefficients m9, røio, and mπ. Other MDCT coefficients may be modified in a similar manner. Indeed, for any MDCT coefficient, the maximum change is limited by the upper and lower limits of its mantissa integer
value Nk . Referring to FIG. 6, for example, the table 600 indicates lower limit and upper
limit values of -0.9333 to +0.9333. As a result, the local masking energy was used to determine the code magnitude, and code symbols may be embedded into an audio signal by modifying eight successive frames of the compressed digital data stream without performing decompression of MDCT coefficients during the encoding process of the methods and apparatus described herein.
[0068] FIG. 14 is a block diagram of an example processor system 2000 that may used to implement the methods and apparatus disclosed herein. The processor system 2000 may be a desktop computer, a laptop computer, a notebook computer, a personal digital assistant (PDA), a server, an Internet appliance or any other type of computing device.
[0069] The processor system 2000 illustrated in FIG. 14 includes a chipset 2010, which includes a memory controller 2012 and an input/output (I/O) controller 2014. As is well known, a chipset typically provides memory and I/O management functions, as well as a plurality of general purpose and/or special purpose registers, timers, etc. that are accessible or used by a processor 2020. The processor 2020 is implemented using one or more processors. In the alternative, other processing technology may be used to implement the processor 2020. The processor 2020 includes a cache 2022, which may be implemented using a first-level unified cache (LI), a second-level unified cache (L2), a third-level unified cache (L3), and/or any other suitable structures to store data. [0070] As is conventional, the memory controller 2012 performs functions that enable the processor 2020 to access and communicate with a main memory 2030 including a volatile memory 2032 and a non-volatile memory 2034 via a bus 2040. The volatile memory 2032 may be implemented by Synchronous Dynamic Random Access Memory (SDRAM), Dynamic Random Access Memory (DRAM), RAMBUS Dynamic Random Access Memory (RDRAM), and/or any other type of random access memory device. The
non-volatile memory 2034 may be implemented using flash memory, Read Only Memory
(ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), and/or any other desired type of memory device.
[0071] The processor system 2000 also includes an interface circuit 2050 that is coupled to the bus 2040. The interface circuit 2050 may "be implemented using any type of well known interface standard such as an Ethernet interface, a universal serial bus
(USB), a third generation input/output interface (3GIO) interface, and/or any other suitable type of interface.
[0072] One or more input devices 2060 are connected to the interface circuit 2050.
The input device(s) 2060 permit a user to enter data and commands into the processor
2020. For example, the input device(s) 2060 may be implemented by a keyboard, a mouse, a touch-sensitive display, a track pad, a track ball, an isopoint, and/or a voice recognition system.
[0073] One or more output devices 2070 are also connected to the interface circuit
2050. For example, the output device(s) 2070 may be implemented by media presentation devices (e.g., a light emitting display (LED), a liquid crystal display (LCD), a cathode ray tube (CRT) display, a printer and/or speakers). The interface circuit 2050, thus, typically includes, among other things, a graphics driver card.
[0074] The processor system 2000 also includes one or more mass storage devices
2080 to store software and data. Examples of such mass storage device(s) 2080 include floppy disks and drives, hard disk drives, compact disks and drives, and digital versatile disks (DVD) and drives.
[0075] The interface circuit 2050 also includes a communication device such as a modem or a network interface card to facilitate exchange of data with external computers via a network. The communication link between the processor system 2000 and the
network may be any type of network connection such as an Ethernet connection, a digital subscriber line (DSL), a telephone line, a cellular telephone system, a coaxial cable, etc. [0076] Access to the input device(s) 2060, the output device(s) 2070, the mass storage device(s) 2080 and/or the network is typically controlled by the I/O controller 2014 in a conventional manner. In particular, the I/O controller 2014 performs functions that enable the processor 2020 to communicate with the input device(s) 2060, the output device(s) 2070, the mass storage device(s) 2080 and/or the network via the bus 2040 and the interface circuit 2050.
[0077] While the components shown in FIG. 14 are depicted as separate blocks within the processor system 2000, the functions performed by some of these blocks may be integrated within a single semiconductor circuit or may be implemented using two or more separate integrated circuits. For example, although the memory controller 2012 and the I/O controller 2014 are depicted as separate blocks within the chipset 2010, the memory controller 2012 and the I/O controller 2014 may be integrated within a single semiconductor circuit.
[0078] The methods and apparatus disclosed herein are particularly well suited for use with data streams implemented in accordance with the AC-3 standard. However, the methods and apparatus disclosed herein may be applied to other digital audio coding techniques.
[0079] In addition, while this disclosure is made with respect to example television systems, it should be understood that the disclosed system is readily applicable to many other media systems. Accordingly, while this disclosure describes example systems and processes, the disclosed examples are not the only way to implement such systems. [0080] Although certain example methods, apparatus, and articles of manufacture have been described herein, the scope of coverage of this patent is not limited thereto. On
the contrary, this patent covers all methods, apparatus, and articles of manufacture fairly falling within the scope of the appended claims either literally or under the doctrine of equivalents. For example, although this disclosure describes example systems including, among other components, software executed on hardware, it should be noted that such systems are merely illustrative and should not be considered as limiting. In particular, it is contemplated that any or all of the disclosed hardware and software components could be embodied exclusively in dedicated hardware, exclusively in firmware, exclusively in software or in some combination of hardware, firmware, and/or software.