Classifications

• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
  • G10L19/08—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters
  • G10L19/12—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters the excitation function being a code excitation, e.g. in code excited linear prediction [CELP] vocoders
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
  • G10L19/0204—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders using subband decomposition
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/002—Dynamic bit allocation
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
  • G10L19/0212—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders using orthogonal transformation
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
  • G10L19/032—Quantisation or dequantisation of spectral components
  • G10L19/038—Vector quantisation, e.g. TwinVQ audio
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
  • G10L19/16—Vocoder architecture
  • G10L19/18—Vocoders using multiple modes
  • G10L19/24—Variable rate codecs, e.g. for generating different qualities using a scalable representation such as hierarchical encoding or layered encoding

Definitions

• the present invention relates to a sound data processing.
• This processing is adapted in particular to the transmission and / or storage of digital signals such as audio-frequency signals (speech, music, or other).
• - waveform coding methods such as PCM (for "Coded Pulse Modulation") and ADPCM (for "Pulse Modulation and Adaptive Differential Coding"), also known as “PCM” "and” ADPCM “in English
• PCM Coded Pulse Modulation
• ADPCM Pulse Modulation and Adaptive Differential Coding
• CELP Code Excited Linear Prediction
• a sound signal such as a speech signal can be predicted from its recent past (for example from 8 to 12 samples at 8 kHz) using parameters evaluated on short windows (10 to 20 ms). in this example).
• These short-term prediction parameters representative of the transfer function of the vocal tract (for example to pronounce consonants), are obtained by LPC (for Linear Prediction Coding) analysis methods.
• LPC Linear Prediction Coding
• a longer-term correlation is also used to determine the periodicities of voiced sounds (eg vowels) due to the vibration of the vocal cords. It is therefore a question of determining at least the fundamental frequency of the voiced signal which varies typically from 60 Hz (deep voice) to 600 Hz (high voice) according to the speakers.
• the LTP long-term prediction parameters including the pitch period, represent the fundamental vibration of the speech signal (when it is voiced), while the LPC short-term prediction parameters represent the spectral envelope. of this signal.
• all of these LPC and LTP parameters thus resulting from a speech coding, can be transmitted in blocks to a peer decoder, via one or more telecommunication networks, to then restore the initial speech signal.
• the encoder In conventional speech coding, the encoder generates a fixed rate bit stream. This flow constraint simplifies the implementation and use of the encoder and decoder. Examples of such systems are the ITU-T G.711 64 kbit / s standard encoding, the ITU-T G.729 8 kbit / s standard encoding, or the 12.2 kbit / s GSM-EFR encoding.
• variable rate bit stream In some applications (such as mobile telephony or VoIP for "Internet Protocol"), it is best to generate a variable rate bit stream. Flow values are taken in a predefined set. Such a coding technique, called “multi-rate” is therefore more flexible than a fixed rate coding technique.
• multi-mode coding controlled by the source and / or the channel, implemented in particular in 3GPP AMR-NB, 3GPP AMR-WB or 3GPP2 VMR-WB coders, the hierarchical coding (or "scalable” coding) that generates a so-called “hierarchical” bitstream because it comprises a core rate and one or more layer (s) enhancement (the G.722 standardized coding at 48, 56 and 64 kbit / s is typically bitrate scalable, while the ITU-T G.729.1 and MPEG-4 CELP codecs are scalable in both bit rate and bandwidth), the multi-description coding described in particular in: "A multiple description speech coder based on AMR-WB for mobile ad hoc networks", H. Dong, A. Gersho, JD Gibson, V. Cuperman, ICASSP, p. 277-280, vol. 1 ( May 2004).
• Hierarchical coding having the capacity to provide varied bit rates, is described below by distributing the information relating to an audio signal to be coded in hierarchical subsets, so that this information can be used in order of importance. in terms of audio rendering quality.
• the criterion taken into account for determining the order is a criterion for optimizing (or rather reducing) the quality of the coded audio signal.
• Hierarchical coding is particularly suited to transmission over heterogeneous networks or having variable available rates over time, or to transmission to terminals with varying capacities.
• the bit stream includes a base layer and one or more enhancement layers.
• the base layer is generated by a low-rate (fixed) codec, known as a "core coded", which guarantees the minimum quality of the coding. This layer must be received by the decoder to maintain an acceptable level of quality. Improvement layers are used to improve quality. However, they may not all be received by the decoder.
• the main advantage of hierarchical coding is that it allows an adaptation of the bit rate simply by "truncation of the bit stream".
• the number of layers i.e., the number of possible truncations of the bitstream
• the number of layers defines the granularity of the coding.
• scalable bandwidth and scalability encoding techniques are described below, with a CELP core-type coder, in a telephone band, and one or more broadband enhancement layer (s).
• An example of such systems is given in the ITU-T G.729.1 8-32 kbit / s fine grain standard.
• the G.729.1 coding / decoding algorithm is summarized below.
• the G.729.1 encoder is an extension of the ITU-T G.729 coder. It is a modified G.729 heart-shaped hierarchical encoder producing a bandwidth ranging from narrow band (50-4000 Hz) to wide band (50-7000 Hz) at a rate of 8 to 32 kbit / s for conversational services. This codec is compatible with existing VoIP devices (most of which are equipped according to G.729). Finally, it should be noted that G.729.1 was approved in May 2006.
• the G.729.1 coder is shown schematically in FIG. 1.
• the broadband input signal s wb sampled at 16 kHz, is first broken down into two subbands by QMF (for "Quadrature Mirror Filter") filtering.
• the low band (0-4000 Hz) is obtained by LP low-pass filtering (block 100) and decimation (block 101), and the high band (4000-8000 Hz) by HP high-pass filtering (block 102) and decimation (block 103).
• the LP and HP filters are of length 64.
• the low band is pretreated with a high-pass filter eliminating the components below 50 Hz (block 104), to obtain the signal s LB , before CELP coding in narrow band (block 105) at 8 and 12 kbit / s.
• This high-pass filtering takes into account that the Useful band is defined as covering the range 50-7000 Hz.
• the narrow-band CELP coding is a cascaded CELP coding comprising as a first stage a modified G.729 coding without pre-processing filter and as a second stage an additional fixed CELP dictionary.
• the high band is first pretreated (block 106) to compensate for the folding due to the high-pass filter (block 102) combined with the decimation (block 103).
• the high band is then filtered by a low pass filter (block 107) eliminating the components between 3000 and 4000 Hz from the high band (i.e. the components between 7000 and 8000 Hz in the original signal) to obtain the signal s HB .
• a band extension (block 108) is then performed.
• the error signal d LB of the low band is calculated (block 109) from the output of the CELP coder (block 105) and a predictive coding by transform (for example of type
• TDAC for "Time Domain Aliasing Cancellation" in the G.729.1 standard) is carried out at block 110.
• the TDAC encoding is applied to both the error signal on the band. bass and the filtered signal on the high band.
• Additional parameters can be transmitted by the block 111 to a homologous decoder, this block 111 performing a so-called “FEC” treatment for "Frame Erasure Concealment", in order to reconstitute possible erased frames.
• the different bit streams generated by the coding blocks 105, 108, 110 and 111 are finally multiplexed and structured into a hierarchical bit stream in the multiplexing block 112.
• the coding is performed by 20 ms sample blocks (or frames). 320 samples per frame.
• the G.729.1 codec therefore has a three-step coding architecture comprising:
• the homologous decoder according to the G.729.1 standard is illustrated in FIG. 2.
• the bits describing each 20 ms frame are demultiplexed in the block 200.
• the bit stream of the 8 and 12 kbit / s layers is used by the CELP decoder (block 201) to generate the narrow-band synthesis (0-4000 Hz).
• the portion of the bit stream associated with the 14 kbit / s layer is decoded by the tape extension module (block 202).
• the portion of the bit stream associated with data rates greater than 14 kbit / s is decoded by the TDAC module (block 203).
• Pre-echo and post-echo processing is performed by blocks 204 and 207 as well as enrichment (block 205) and aftertreatment of the low band (block 206).
• the extended band output signal s wb sampled at 16 kHz, is obtained via the QMF synthesis filter bank (blocks 209, 210, 211, 212 and 213) integrating the inverse folding (block 208).
• the TDAC type transform coding in the G.729.1 encoder is illustrated in FIG.
• the filter W LB (z) (block 300) is a perceptual weighting filter, with gain compensation, applied to the low band error signal d LB.
• MDCT transforms are then calculated (block 301 and 302) to obtain: - the MDCT spectrum D ⁇ 3 of the difference signal, perceptually filtered, and the MDCT spectrum S HB of the original signal of the high band.
• MDCT transforms (blocks 301 and 302) apply to 20 ms of sampled signal at 8 kHz (160 coefficients).
• the spectrum Y (k) from the block 303 of fusion thus comprises 2 x 160, or 320 coefficients. It is defined as follows:
• This spectrum is divided into eighteen sub-bands, a sub-band j being assigned a number of coefficients noted nb_coef (j).
• the subband splitting is specified in Table 1 below.
• a sub-band j comprises the coefficients Y (k) with sb-bound (j) ⁇ k ⁇ sb-bound (j + 1).
• the spectral envelope JlOg-ZmS (J) J _ o ⁇ is calculated in block 304 according to the formula:
• This value rms _index (j) is transmitted to bit allocation block 306.
• two types of coding may be chosen according to a given criterion, and, more precisely, the rms values _index (j): may be coded by "differential Huffman" coding, - or may be coded by natural binary coding .
• a bit (0 or 1) is transmitted to the decoder to indicate the encoding mode that has been chosen.
• the number of bits allocated to each sub-band for its quantization is determined in block 306 from the quantized spectral envelope from block 305.
• the allocation of the bits performed minimizes the squared error while respecting the constraint of a number of integer bits allocated per subband and a maximum number of bits not to be exceeded.
• the spectral content of the subbands is then encoded by spherical vector quantization (block 307).
• the step of TDAC-type transform decoding in the G.729.1 decoder is illustrated in FIG. 4.
• the decoded spectral envelope (block 401) makes it possible to recover the allocation.
• each of the subbands is found by inverse spherical vector quantization (block 403).
• the non-transmitted sub-bands, due to a lack of "budget" of bits, are extrapolated (block 404) from the MDCT transform of the signal at the output of the band extension block (block 202 of FIG. 2).
• the spectrum MDCT is separated into two (block 407): with 160 first coefficients corresponding to the spectrum D ⁇ B of a decoded difference signal in low band, perceptually filtered, and 160 subsequent coefficients corresponding to the spectrum S HB of the original decoded signal in high band.
• IMDCT inverse MDCT transform time signals
• W 18 (Z) 1 the inverse perceptual weighting filter
• W 18 (Z) 1 the inverse perceptual weighting
• Table 2 Possible values of number of bits allocated in TDAC subbands.
• nbit (j) arg rR m, in nb_ coef (j) x (ip (j) -A) - r
• ⁇ is a parameter optimized by dichotomy.
• the TDAC coding uses the perceptual weighting W LB (z) filter in the low band (block 300), as indicated above.
• perceptual weighting filtering allows you to format the coding noise.
• the principle of this filtering is to exploit the fact that it is possible to inject more noise in the frequency zones where the original signal has a high energy.
• the most common perceptual weighting filters used in narrow-band CELP coding are of the form ⁇ (z / ⁇ 1) / ⁇ (z / ⁇ 2) where 0 ⁇ 2 ⁇ l ⁇ 1 and ⁇ (z) represents a prediction spectrum linear (LPC).
• CELP coding synthesis analysis thus, it amounts to minimizing the quadratic error in a perceptually weighted signal domain by this type of filter.
• the W LB (z) filter is defined as:
• the fac factor makes it possible to ensure at the junction of the low and high bands (4 kHz) a gain of the filter at 1 to 4 kHz. It is important to note that in the G.729.1 TDAC coding, the coding is based on an energetic criterion only.
• the TDAC encoder jointly processes: the difference signal, between the original low band and the CELP synthesis, perceptually filtered by a ⁇ (z / ⁇ 1) / ⁇ (z / ⁇ 2) compensated filter. gain (ensuring spectral continuity), and - the high band which contains the original high band signal.
• the low band signal corresponds to the frequencies 50 Hz-4 kHz, while the high band signal corresponds to the frequencies 4-7 kHz.
• the joint coding of these two signals is carried out in the MDCT domain according to the criterion of the quadratic error.
• the high band is coded according to energy criteria, which is suboptimal (in the "perceptual" sense of the term). More generally, the case of multi-band coding may be considered, a perceptual weighting filter being applied to the signal of at least one band in the time domain, and the set of subbands being coded together. by transform coding. If we want to apply the perceptual weighting in the frequency domain, then there is the problem of continuity and homogeneity of the spectra between subbands.
• the present invention improves the situation.
• the method comprises: a determination of at least one frequency masking threshold to be applied on the second subband and normalizing said masking threshold to provide spectral continuity between said first and second subbands.
• the present invention therefore proposes to calculate a frequency perceptual weighting, using a masking threshold, on only a part of the frequency band (at least on the "second subband” mentioned above) and to ensure spectral continuity with at least another frequency band (at least the aforementioned "first sub-band”) by normalizing the masking threshold on the spectrum covering these two frequency bands.
• the allocation of the bits for the second sub-band at least is determined furthermore according to a standardized masking curve calculation, applied at least to the second sub-band.
• the application of the invention makes it possible to assign the bits to the sub-bands which require the most bits according to a perceptual criterion.
• perceptual frequency weighting is then applied by masking a part of the audio band, so as to improve the audio quality by optimizing in particular the distribution of bits between subbands according to criteria. perceptual.
• the transformed signal in the second subband is weighted by a factor proportional to the square root of the normalized masking threshold for the second subband.
• the normalized masking threshold is not used for the allocation of the bits to the subbands as in the first application mode above, but it can advantageously be used to directly weight the signal of the second sub-band at least in the transformed domain.
• the present invention is advantageously, but not exclusively, applied to a TDAC-type transform coding in a global encoder according to the G.729.1 standard, the first subband being included in a low frequency band, whereas the second subband is included in a low frequency band, while the second subband is included in a low frequency band.
• -band is included in a high frequency band that can extend up to 7000 Hz, or even more (typically up to 14 kHz) in band extension.
• the application of the invention may then consist in providing a perceptual weighting for the high band while ensuring spectral continuity with the low band.
• the first subband then comprises a signal resulting from a core coding of the hierarchical coder
• the second subband comprises an original signal
• the signal from the core coding can be perceptually weighted and the implementation of the invention is advantageous in the sense that the entire spectral band can finally be perceptually weighted.
• the signal from the core coding may be a signal representative of a difference between an original signal and a synthesis of this original signal (called “difference signal” or “error signal”). .
• difference signal or "error signal”
• the present invention also relates to a decoding method, homologous to the coding method defined above, in which at least one first and one second subband, adjacent, are decoded by transform.
• the decoding method then comprises: a determination of at least one frequency masking threshold to be applied on the second subband, starting from a decoded spectral envelope, and a normalization of said masking threshold to provide spectral continuity between said first and second subbands.
• a first decoding application mode homologous to the first application mode of the coding defined above, aims at the allocation of bits to the decoding and a number of bits to be allocated to each subband is determined from a decoding spectral envelope.
• the bit allocation for the at least second subband is further determined according to a normalized masking curve calculation applied at least to the second subband.
• a second method of applying decoding within the meaning of the invention consists in weighting the transformed signal in the second subband by the square root of the normalized masking threshold. This embodiment will be described in detail with reference to FIG.
• FIG. 5 illustrates an advantageous spreading function for masking in the sense of the invention
• FIG. 6 illustrates, for comparison with FIG. 3, the structure of a TDAC encoding using a masking curve calculation 606 for the allocation of bits according to a first embodiment of the invention
• FIG. 7 illustrates, for comparison with FIG. 4, the structure of a homologous TDAC decoding of FIG. 6, using a curve calculation 702 according to the first embodiment of the invention
• FIG. 8 illustrates a normalization of the masking curve, in a first embodiment where the sampling frequency is 16 kHz and the masking of the invention applied for the high band 4-7 kHz
• FIG. 9A illustrates the structure of a modified TDAC encoding, with directly weighting of the signal in the high frequencies 4-7 kHz in a second embodiment of the invention, and coding of the standardized masking threshold
• FIG. 9B illustrates the structure of a TDAC encoding in a variant of the second application mode illustrated in FIG. 9A, here with a coding of the spectral envelope
• FIG. 10A illustrates the structure of a homologous TDAC decoding of FIG. 9A, according to the second embodiment application of the invention
• FIG. 10A illustrates the structure of a homologous TDAC decoding of FIG. 9A, according to the second embodiment application of the invention
• FIG. 1B illustrates the structure of a homologous TDAC decoding of FIG. 9B, according to the second embodiment of the invention, with here a calculation of the decoding masking threshold
• FIG. 11 illustrates the normalization of the curve. in a second embodiment of the invention where the sampling frequency is 32 kHz and the masking of the invention applied for the high band widened from 4 to 14 kHz
• FIG. 12 illustrates the spectral power, at the output of the CELP coding, of the difference signal D LB (in solid lines) and of the original signal S LB (in dashed lines).
• the invention provides an improvement to the perceptual weighting performed in the transform coder by exploiting the masking effect known as "simultaneous masking" or "frequency masking".
• This property corresponds to the modification of the hearing threshold in the presence of a so-called “masking” sound. This phenomenon is observed typically when, for example, one tries to hold a discussion with ambient noise, for example in the street and that the noise of a vehicle comes to "hide” the voice of a speaker.
• an approximate masking threshold is calculated for each spectrum line. This threshold is the one above which the line concerned is supposed to be audible.
• the masking threshold is calculated from the convolution of the signal spectrum with a spreading function B (v) modeling the masking effect of a sound (sinusoid or filtered white noise) by another sound (sinusoid or noise filtered white).
• FIG. 5 An example of such a spreading function is shown in FIG. 5. This function is defined in a frequency domain whose unit is Bark. The frequency scale is representative of the frequency sensitivity of the ear. A usual approximation of the conversion of a frequency / in Hertz, in "frequencies" noted ⁇ (in Barks), is given by the following relation:
• the calculation of the masking threshold is performed by subband rather than by line.
• the threshold thus obtained is used to perceptually weight each of the subbands.
• the allocation of the bits is thus performed, not by minimizing the square error but by minimizing the "mask-to-mask noise" ratio, the aim being to shape the coding noise so that it is inaudible ( below the masking threshold).
• the spreading function may be a function of the level of the line and / or the frequency of the masking line. Detection of "peaks" can also be implemented.
• An application of the invention described hereinafter makes it possible to improve the TDAC coding of the encoder according to the G.729.1 standard, in particular by applying a perceptual weighting of the high band (4 to 7 kHz) while ensuring the continuity spectral between low and high bands for a satisfactory and joint coding of these two bands.
• the input signal is sampled at 16 kHz, bandwidth 50 Hz to 7 kHz.
• the encoder always operates at the maximum rate of 32 kbit / s, while the decoder can receive the core (8 kbit / s), as well as one or more enhancement layers (12 to 32 kbit / s per step). 2 kbit / s), as in G.729.1.
• the coding and decoding have the same architecture as that shown in FIGS. 1 and 2. Here, only blocks 110 and 203 are modified as described in FIGS. 6 and 7.
• the modified TDAC coder is identical to that of FIG. 3, except that the allocation of bits following the quadratic error (block 306) is now replaced by a masking curve calculation and a modified bit allocation (blocks 606 and 607), the invention forming part of the calculation of the masking curve (block 606) and its use in the allocation of bits (block 607).
• the modified TDAC decoder is shown in FIG. 7 in this first embodiment.
• This decoder is identical to that of FIG. 4, except that the allocation of bits following the quadratic error (block 402) is replaced by a masking curve calculation and a modified bit allocation (blocks 702 and 703). .
• the invention relates to blocks 702 and 703.
• the masking threshold M (J) of the sub-band j is defined by the convolution of the energy envelope ⁇ 2 (J) -rms -q (j) 2 ⁇ nb-coef (j), by a function spreading B (v).
• this masking is performed only on the high band of the signal, with:
• the masking threshold M (J) for a sub-band j is therefore defined by a convolution between:
• FIG. 5 An advantageous spreading function is that shown in FIG. 5. It is a triangular function whose first slope is + 27dB / Bark and -10dB / Bark for the second. This representation of the spreading function allows the iterative calculation of the following masking curve:
• a 1 (J) and A 2 (J) can be pre-calculated and stored.
• a first embodiment of the invention is described below for the allocation of bits in a hierarchical coder such as the G.729.1 encoder.
• the bit allocation criterion is based here on the signal-to-mask ratio given by The low band is already filtered perceptually, the application of the masking threshold is limited to the high band. In order to ensure the spectral continuity between the low and high band spectrum weighted by the masking threshold and to avoid biasing the bit allocation, the masking threshold is normalized by its value on the last subband of the low band.
• normfac log ; ⁇ a 2 U) XB (V 9 - V 1 )
• log_ mask (j) log 2 [M (J)) - normfac.
• the second line of the brace for the calculation of the perceptual importance is an expression of the implementation of the invention according to this first application to the allocation of bits in a transform coding as an upper layer a hierarchical coder.
• An illustration of the standardization of the masking threshold is given in FIG. 8, showing the connection of the high band on which the masking (4-7 kHz) is applied to the low band (0-4 kHz).
• a 0 1 is obtained by dichotomy as in the G.729.1 standard.
• the standardization of the masking threshold can be rather carried out from the value of the band.
• normfac log 2 ⁇ ⁇ 2 (j) x ⁇ (v 10 -v ; )
• the masking threshold can be calculated over the entire frequency band, with:
• the masking threshold is then applied only to the high band after normalization of the masking threshold by its value on the last subband of the low band:
• normfac log 2 B (V 10 - V 1 )
• these relations giving the normalization factor normfac or the masking threshold M (j) can be generalized to any number of sub-bands (different, in total, from eighteen) in the high band (with a different number of eight), as in low band (with a different number of ten).
• the normalized masking threshold is not used to weight the energy in the definition of the perceptual importance, as in the first embodiment described above, but it serves to directly weight the high band signal before TDAC coding.
• FIGS. 9A and 9B This second embodiment is illustrated in FIGS. 9A (for encoding) and 10A (for decoding).
• FIGS. 9B A variant of this second mode, which is the object of the present invention, in particular for the decoding performed, is illustrated in FIGS. 9B (for encoding) and 10B (for decoding).
• FIGS. 9A and 9B the spectrum Y (k) from block 903 is divided into eighteen subbands and the spectral envelope is calculated (block 904) as previously described.
• the masking threshold is calculated (block 905 of FIG. 9A and block 906b of FIG. 9B) from the unquantized spectral envelope.
• information representative of the weighting is directly coded by the masking threshold M (J), rather than encoding the spectral envelope.
• This coding is performed by algebraic quantization according to the quadratic error, as described in the document Ragot et al: "Low-complexity multi-rate lattice vector quantization with application to wideband TCX speech coding at 32 kbit / s", S. Ragot, B. Bessette, and R. Lefebvre, Proceedings ICASSP - Montreal (Canada), Pages: 501-504 , vol. 1 (2004).
• This gain-form type quantization method is implemented in particular in the 3GPP AMR-WB + standard.
• the peer decoder is shown in Figure 10A.
• the scaling factors sf _q (j), j - 0, - - -, 17, are decoded in the block 1001.
• the block 1002 is then carried out as described in the document Ragot et al. supra.
• This second embodiment may be particularly advantageous especially in an implementation according to the standard 3 GPP-AMR-WB + which is presented as the preferred context of the document Ragot et al. supra.
• the coded information remains the envelope of FIG. energy (rather than the masking threshold itself as in Figs. 9A and 10A).
• the masking threshold is calculated and normalized (block 906b of FIG. 9B) from the coded spectral envelope (block 905b).
• the masking threshold is calculated and normalized (block 1011b of FIG. 10B) from the decoded spectral envelope (block 1001b), the decoding of the envelope making it possible to perform a level adjustment (block 1010b of FIG. 10B) from the values quantified rms_q (j).
• a masking threshold is calculated for each sub-band, at least for the sub-bands of the high frequency band, this masking threshold being normalized to ensure spectral continuity between the subbands concerned.
• the calculation of the masking threshold is particularly advantageous when the signal to be coded is not tonal, in the first mode, as in the second embodiment, described above.
• the application of the spreading function B (v) results in a masking threshold very close to a tone a little more spread out in frequencies.
• the allocation criterion minimizing the masked coding noise ratio then gives a bit of bit allocation.
• the same is true for the direct weighting of the high band signal according to the second embodiment. It is therefore preferred, for a tonal signal, to use a bit allocation according to energy criteria.
• the invention is applied only if the signal to be encoded is not tonal.
• the bit relating to the mode of the coding of the spectral envelope indicates a "differential Huffman" mode or a "natural direct binary” mode.
• This mode bit can be interpreted as a tone detection, since, in general, a tonal signal leads to envelope coding by the "natural direct binary” mode, while most non-tonal signals, having a spectral dynamic more limited, lead to envelope coding by the "Differential Huffman" mode.
• the module 904 of FIG. 9A can, by calculating the spectral envelope, determine whether the signal is tonal or not and thus Block 905 is bypassed if yes.
• the module 904 can make it possible to determine whether the signal is tonal or not and thus bypass the block 907 in the affirmative.
• Figure 11 generalizes the normalization of the masking curve (described in Figure 8) in the case of super-wide band coding.
• the signals in this embodiment are sampled at a frequency of 32 kHz (instead of 16 kHz) for a useful band of 50 Hz - 14 kHz.
• the masking curve log 2 [M (J)] is then defined at least for the sub-bands ranging from 7 to 14 kHz.
• the spectrum covering the band 50 Hz - 14 kHz is coded by subbands and the allocation of bits to each subband is made from the spectral envelope as in the G.729.1 encoder.
• a partial masking threshold can be calculated as previously described.
• the standardization of the masking threshold is thus generalized to the case where the high band has more subbands or covers a wider frequency range than in the G.729.1 standard.
• a first transform T1 is applied to the time weighted difference signal.
• a second transform T2 is applied to the signal on the first high band between 4 and 7 kHz and a third transform T3 is applied to the signal on the second high band between 7 and 14 kHz.
• the invention is not limited to signals sampled at 16kHz. Its implementation is particularly advantageous also for signals sampled at higher frequencies, such as for the extension of the G.729.1 encoder to signals sampled not at 16 kHz but at 32 kHz, as described above. If the TDAC coding is generalized to such a frequency band (50 Hz - 14 kHz instead of 50 Hz - 7 kHz currently), the advantage provided by the invention will be really major.
• the invention also aims to improve the TDAC coding, in particular by applying a perceptual weighting of the high-bandwidth (4-14 kHz) while ensuring the spectral continuity between bands, this criterion being important for a joint coding of the band.
• first low band and the second high and extended band up to 14 kHz.
• the hierarchical coder is implemented with a heart coder in a first frequency band, and the error signal associated with this heart coder is directly transformed, without perceptual weighting in this first frequency band, to be coded. together with the transformed signal of a second frequency band.
• the original signal can be sampled at 16 kHz and decomposed into two frequency bands (from 0 to
• the encoder can typically be, in such an embodiment, an encoder according to the standard
• the transform coding is then performed on: the signal difference between the original signal and the synthesis G.711 in the first frequency band (0-4000 Hz), and the original signal, perceptually weighted in the frequency domain according to the invention, in a second frequency band (4000-8000 Hz).
• the perceptual weighting in the low band is not necessary for the application of the invention.
• the original signal is sampled at 32 kHz and decomposed into two frequency bands (0 to 8000 Hz and 8000 to 16000 Hz) by an appropriate filter bank, QMF type.
• the encoder can be here an encoder according to the G.722 standard (ADPCM compression in two sub-bands), and the transform coding is performed on: the signal difference between the original signal and the synthesis signal G.122 in the first frequency band (0-8000 Hz), and the original signal, which is still weighted perceptually according to the invention in a frequency domain restricted to the second frequency band (8000-16000 Hz).
• the present invention also relates to a first computer program, stored in a memory of an encoder of a telecommunication terminal and / or stored on a memory medium intended to cooperate with a reader of said encoder.
• This first program then comprises instructions for implementing the coding method defined above, when these instructions are executed by an encoder processor.
• the present invention also relates to an encoder comprising at least one memory storing this first computer program.
• FIGS. 6, 9A and 9B may constitute flowcharts of this first computer program, or further illustrate the structure of such an encoder, according to distinct embodiments and variants.
• the present invention also relates to a second computer program, stored in a memory of a decoder of a telecommunication terminal and / or stored on a storage medium intended to cooperate with a reader of said decoder.
• This second program then comprises instructions for implementing the decoding method defined above, when these instructions are executed by a processor of the decoder.
• the present invention also relates to a decoder comprising at least one memory storing this second computer program.
• FIGS. 7, 10A, 10B can constitute flowcharts of this second computer program, or further illustrate the structure of such a decoder, according to different embodiments and variants.

Applications Claiming Priority (2)

Publications (2)

Family

ID=38477199

Family Applications (1)

Country Status (10)

Families Citing this family (46)

Family Cites Families (16)

• 2007
  • 2007-02-02 FR FR0700747A patent/FR2912249A1/fr not_active Withdrawn
• 2008
  • 2008-01-30 WO PCT/FR2008/050150 patent/WO2008104663A1/fr active Application Filing
  • 2008-01-30 EP EP08762010A patent/EP2115741B1/de active Active
  • 2008-01-30 CN CN2008800066533A patent/CN101622661B/zh active Active
  • 2008-01-30 JP JP2009547737A patent/JP5357055B2/ja active Active
  • 2008-01-30 DE DE602008001718T patent/DE602008001718D1/de active Active
  • 2008-01-30 US US12/524,774 patent/US8543389B2/en active Active
  • 2008-01-30 AT AT08762010T patent/ATE473504T1/de not_active IP Right Cessation
  • 2008-01-30 ES ES08762010T patent/ES2347850T3/es active Active
  • 2008-01-30 KR KR1020097016113A patent/KR101425944B1/ko active IP Right Grant

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events