ES2351935T3 - PROCEDURE AND APPARATUS FOR VECTOR QUANTIFICATION OF A SPECTRAL ENVELOPE REPRESENTATION. - Google Patents

Abstract

A method for signal processing, said method comprising: encoding a first frame and a second frame of a voice signal to produce corresponding first and second vectors, in which the first vector represents a spectral envelope of the voice signal during the first frame and the second vector represents a spectral envelope of the voice signal during the second frame; generating a first quantified vector, including said generation quantifying a third vector that is based on the first vector, calculating a quantization error of the first quantized vector; calculate a fourth vector, said calculation including adding a scale version of the quantization error to the second vector; and quantify the fourth vector.

Description

FIELD OF THE INVENTION

The present invention is about signal processing.

BACKGROUND

A voice encoder sends a spectral envelope characterization of a voice signal to a decoder in the form of a line spectral frequency vector (LSF) or a similar representation. For efficient transmission, these LSFs are quantified.

"Error Spectrum Shaping and Vector Quantization," by Jon Dattorro and Christine Law, published in October 1997, demonstrates in the form of an experimental report that the return information of the truncation error is not successful in quantifying still images.

SUMMARY

In one aspect of the invention, there is provided a method for signal processing as set forth in claim 1.

In a further aspect of the invention, an apparatus is provided for signal processing as set forth in claim 8. A computer-readable medium is also provided as set forth in claim 18.

A quantifier, according to one embodiment, is configured to quantify a flattened value of an input value (such as a vector of line spectral frequencies or a portion thereof) to produce a corresponding output value, based on the flattened value on a scale factor and in a quantization error of a previous output value.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGURE 1a shows a block diagram of a voice E100 encoder according to one embodiment.

FIGURE 1b shows a block diagram of a voice E200 encoder.

FIGURE 2 shows an example of a one-dimensional correspondence normally carried out by means of a scalar quantifier.

FIGURE 3 shows a simple example of a multidimensional correspondence as carried out by means of a vector quantifier.

FIGURE 4a shows an example of a one-dimensional signal, and FIGURE 4b shows an example of a version of this signal after being quantified.

FIGURE 4c shows an example of the signal of FIGURE 4a as quantified by a quantifier 230b, as shown in FIGURE 6.

FIGURE 4d shows an example of the signal of FIGURE 4a as quantified by a quantifier 230a, as shown in FIGURE 5.

FIGURE 5 shows a block diagram of an implementation 230a of a quantizer 230 according to one embodiment.

FIGURE 6 shows a block diagram of an implementation 230b of a quantizer 230 according to one embodiment.

FIGURE 7a shows an example of a logarithmic amplitude plot as a function of the frequency for a voice signal.

FIGURE 7b shows a block diagram of a basic linear prediction coding system.

FIGURE 8 shows a block diagram of an A122 implementation of a narrowband encoder A 120.

FIGURE 9 shows a block diagram of an implementation B112 of a narrow band decoder B110.

FIGURE 10a is a block diagram of a broadband voice encoder A100.

FIGURE 10b is a block diagram of an A102 implementation of a broadband voice encoder A 100.

FIGURE 11a is a block diagram of a broadband voice decoder B100 corresponding to the broadband voice encoder A100.

FIGURE 11b is an example of a broadband voice decoder B102 corresponding to the broadband voice encoder A102.

DETAILED DESCRIPTION

Due to the quantization error, the reconstructed spectral envelope in the decoder may exhibit excessive fluctuations. These fluctuations can produce an unacceptable "vibrated" quality in the decoded signal. Embodiments include systems, procedures, and apparatus configured to perform high quality broadband voice coding using a quantification of temporal noise formation of spectral envelope parameters. Features include a fixed or adaptable flattening of coefficient representations, such as high band LSF. The particular applications described herein include a broadband voice encoder that combines a narrow band signal with a high band signal.

Unless expressly limited by context, the term "calculate" is used herein to indicate any of its normal meanings, such as calculate, generate, and select from a list of values. When the term "understand" is used in the present invention and in the claims, it does not exclude other elements or operations. The expression "A is based on B" is used to indicate any of its normal meanings, including cases (i) "A is equal to B" and (ii) "A is based on at least B". The term "Internet Protocol" includes version 4, as described in IETF (Internet Engineering Task Force) RFC (Request for Comments) 791, and subsequent versions such as version 6.

A voice encoder can be implemented according to a source-filter model that encodes the input voice signal as a set of parameters that describe a filter. For example, a spectral envelope of a voice signal is characterized by a number of peaks that represent resonances of the vocal apparatus and are called formants. FIGURE 7a shows an example of such a spectral envelope. Most voice encoders encode at least this gross spectral structure as a set of parameters, such as filter coefficients.

FIGURE 1a shows a block diagram of a voice E100 encoder according to one embodiment. As shown in this example, the analysis module can be implemented as a linear prediction coding analysis (LPC) module 210 that encodes the spectral envelope of the voice signal S1 as a set of linear prediction coefficients (LP) (for example, coefficients of an all-pole filter 1 / A (z)). Normally, the analysis module processes the input signal as a series of non-overlapping frames, calculating a new set of coefficients for each frame. In general, the frame period is a period during which the signal can be expected to be locally stationary; A common example is 20 milliseconds (equivalent to 160 samples at an 8 kHz sampling rate). An example of a low-band LPC analysis module (as shown, for example, in FIGURE 8 as the LPC analysis module 210) is configured to calculate a set of ten LP filter coefficients to characterize the formation structure of each 20 millisecond frame of the narrowband signal S20, and an example of a high-band LPC analysis module (as shown, for example, in FIGURE 10a as the high-band encoder A200 ) is configured to calculate a set of six (alternatively, eight) LP filter coefficients to characterize the formative structure of each 20 millisecond frame of the high band signal S30. It is also possible to implement the analysis module to process the input signal as a series of overlapping frames.

The analysis module can be configured to analyze the samples of each frame directly, or the samples can be weighted first according to a function of windows (for example, a Hamming window). The analysis can also be carried out in a window that is larger than the frame, such as a 30 msec window. This window can be symmetric (for example, 5-20-5, so that it includes the 5 milliseconds immediately before and after the 20 millisecond frame) or be asymmetric (for example, 10-20, so that it includes the last 10 milliseconds of the previous plot). Normally, there is an LPC analysis module configured to calculate the LP filter coefficients using a Levinson-Durbin recursion or the Leroux-Gueguen algorithm. In another implementation, the analysis module may be configured to calculate a set of cepstral coefficients for each frame instead of a set of LP filter coefficients.

The rate of transmission of output bits of a speech encoder can be significantly reduced, with a relatively small effect on playback quality, by quantifying the filter parameters. Linear prediction filter coefficients are difficult to quantify effectively and normally the voice encoder establishes a correspondence between them and another representation, such as line spectral pairs (LSP) or line spectral frequencies (LSF), for quantification and / or entropic coding. The voice encoder E100, as shown in FIGURE 1a includes a transform 220 of the LP to LSF filter coefficient configured to transform the set of LP filter coefficients into a corresponding vector of LSF S3. Other one-to-one representations of LP filter coefficients include parcor coefficients; logarithm-area relationship values; immitance spectral pairs (ISP); and immitance spectral frequencies (ISF), which are used in the GSM codec (Global System for Mobile Communications) AMR-WB (Adaptive Multitasking Bandwidth). Normally, a transform between a set of LP filter coefficients and a corresponding set of LSF can be reversed, but the embodiments also include implementations of a speech encoder in which the transform is not reversible without error.

Normally, a voice encoder includes a quantifier configured to quantify the narrowband LSF set (or other coefficient representation) and to output the result of this quantification as the filter parameters. Normally, quantification is carried out using a vector quantifier that encodes the input vector as an index to a corresponding vector entry in a table or in a list of keys. This quantifier can also be configured to perform a classified vector quantification. For example, such a quantifier can be configured to select one of a set of codes based on information that has already been encoded within the same frame (for example, in the low band channel and / or in the high band channel ). Typically, such a technique provides greater coding efficiency at the expense of additional code storage.

FIGURE 1b shows a block diagram of a corresponding voice decoder E200 that includes a reverse quantizer 310 configured to quantify the quantified LSF S3, and a transform 320 of the LSF to LP filter coefficient configured to transform the quantified LSF vector into a set of LP filter coefficients. Normally, a synthesis filter 330 is activated, configured according to the LP filter coefficients, by means of an excitation signal to produce a synthesized reproduction, that is, a decoded voice signal S5, of the voice input signal. The excitation signal may be based on a random noise signal and / or a quantized representation of the residual signal as it is sent by the encoder. In some multiband encoders, such as a broadband voice encoder A100 and a decoder B 100 (as described herein with reference, for example, to FIGURES 10a, b and 11a, b), the excitation signal for One band is derived from the excitation signal for another band.

The quantification of the LSF introduces a random error that is not normally correlated from one frame to the next. This error can cause quantified LSFs to be less flat than unquantified LSFs and can reduce the perceptual quality of the decoded signal. Independent quantification of LSF vectors generally increases the amount of frame-to-frame spectral fluctuation compared to unquantified LSF vectors, and these spectral fluctuations can cause the decoded signal to sound unnatural.

Knagenhjelm and Kleijn proposed a complicated solution, in which a flattening of the unquantified LSF parameters is carried out in the decoder. This reduces spectral fluctuations, but at the expense of an additional delay. The present application describes procedures that use a temporary noise formation on the encoder side, so that spectral fluctuations can be reduced without further delay.

Typically, a quantifier is configured to correspond between an input value and one of a set of differentiated output values. A number of output values are available, so that a correspondence between a range of input values and a single output value is established. Quantification increases the coding efficiency because an index indicating the corresponding output value can be transmitted in fewer bits than the original input value. FIGURE 2 shows an example of a one-dimensional correspondence typically performed by a scalar quantifier.

The quantifier could equally well be a vector quantifier, and LSFs are normally quantified using a vector quantifier. FIGURE 3 shows a simple example of a multidimensional correspondence as carried out by a vector quantifier. In this example, the input space is divided into a number of Voronoi regions (for example, according to a criterion of the nearest neighbor). The quantification establishes a correspondence between each input value and a value that represents the corresponding Voronoi region (usually, the centroid), shown here as a point. In this example, the input space is divided into six regions, so that any input value can be represented by an index that has only six different states.

If the input signal is very flat, it can sometimes occur that the quantized output is much less flat, according to a minimum step between values in the quantization output space. FIGURE 4a shows an example of a flat one-dimensional signal that varies only within a quantification level (only one such level is shown here), and FIGURE 4b shows an example of this signal after quantification. Although the input of FIGURE 4a varies only in a small range, the resulting output in FIGURE 4b contains more abrupt transitions and is much less flat. Such an effect may result in audible damage, and it may be desirable to reduce this effect for LSFs (or other representations of the spectral envelope to be quantified). For example, the quantification performance of LSFs can be improved by incorporating a temporary noise formation.

In a method according to one embodiment, a spectral envelope parameter vector is estimated once for each voice frame (or other block) in the encoder. The parameter vector is quantified for efficient transmission to the decoder. After quantification, the quantization error is stored (defined as the difference between a quantized and unquantified parameter vector). The quantization error of the N-1 frame is reduced by means of a scale factor and added to the frame vector of frame N, before quantifying the parameter vector of the frame N. It may be desirable for the value of the Scaling factor be smaller when the difference between current and previous estimated spectral envelopes is relatively large.

In a procedure according to one embodiment, the LSF quantization error vector for each frame is calculated and multiplied by a scale factor b having a value less than 1.0. Before quantification, the scale quantification error for the previous frame is added to the LSF vector (input value V10). A quantification operation can be described

5

10

fifteen

twenty

25

30

7

of such a procedure by means of an expression, such as the following:

() = Q (() n + by n −1) - s (n −1)]),

yn s [(

in which s (n) is the flattened LSF vector belonging to frame n, and (n) is the quantified LSF vector belonging to frame n, Q (⋅) is a nearest neighbor quantification operation, and b It is the scale factor.

A quantizer 230 according to one embodiment is configured to produce a quantized output value V30 of a flattened value V20 of an input value V10 (for example, an LSF vector), the flattened value V20 based on a scale factor V40 and an error of quantification of a previous output value V30. Such a quantifier can be applied to reduce spectral fluctuations without additional delay. FIGURE 5 shows a block diagram of an implementation 230a of quantizer 230, in which the values that may be particular to this implementation are indicated by index a. In this example, a quantization error is calculated by subtracting the current input value V10 from the current output value V30a as quantified by the inverse quantizer Q20. The error is stored in a delay element DE10. The flattened value V20a is a sum of the current input value V10 and the quantization error of the previous frame as its scale changes (for example, multiplied) by means of the scale factor V40. The quantifier 230a can also be implemented so that the scale factor V40 is applied before storing the quantization error in the delay element DE10 instead.

FIGURE 4d shows an example of a (unquantified) sequence of output values V30a as produced by quantizer 230a in response to the input signal of FIGURE 4a. In this example, the value of the scale factor V40 is set to 0.5. It can be seen that the signal of FIGURE 4d is flatter than the fluctuating signal of FIGURE 4a.

It may be desirable to use a recursive function to calculate the amount reintroduced. For example, the quantization error can be calculated with respect to the current input value instead of with respect to the current flattened value. Such a procedure can be described by means of an expression such as the following:

n Qs () n [

y () = [()], n sn = x () + by (n −1) - s (n −1)],

where x (n) is the input LSF vector belonging to frame n.

FIGURE 6 shows a block diagram of an implementation 230b of quantizer 230, in which the values that may be particular to this implementation are indicated by index b. In this example, a quantization error is calculated by subtracting the current value of the flattened value V20b from the current output value V30b as quantified by the inverse quantizer Q20. The error is stored in delay element DE10. The flattened value V20b is a sum of the current input value V10 and the quantization error of the previous frame as its scale changes (for example, multiplied) by means of the scale factor V40. Quantifier 230b can also be implemented so that the scale factor V40 is applied before storing the quantization error in the delay element DE10 instead. It is also possible to use different V40 scale factor values in implementation 230a as opposed to implementation 230b.

FIGURE 4c shows an example of a (unquantified) sequence of output values V30b as produced by quantizer 230b in response to the input signal of FIGURE 4a. In this example, the value of the scale factor V40 is set to 0.5. It can be seen that the signal of FIGURE 4c is flatter than the fluctuating signal of FIGURE 4a.

It is noted that embodiments can be implemented as shown herein by replacing or augmenting an existing quantifier Q10 according to an arrangement as shown in FIGURE 5 or 6. For example, quantifier Q10 can be implemented as a vector quantifier. predictive, a multistage quantifier, a quantifier of the separator vector, or according to any other LSF quantification model.

In one example, the scale factor value is set to a desired value between 0 and 1. Alternatively, it may be desirable to adjust the scale factor value dynamically. For example, it may be desirable to adjust the value of the scale factor depending on the degree of fluctuation already present in the unquantified LSF vectors. When the difference between the current and previous LSF vectors is large, the scale factor is close to zero and, as a result, almost no noise is formed. When the current LSF vector differs little from the previous one, the scale factor is close to 1.0. Thus, transitions in the spectral envelope can be retained over time, minimizing spectral distortion when the voice signal changes, while spectral fluctuations can be reduced when the voice signal is relatively constant from one frame to the next. .

The scale factor value can be made proportional to the distance between consecutive LSFs, and any of several distances between vectors can be used to determine the change between LSF. Normally, the Euclidean norm is used, but others that can be used include the distance of Manhattan (norm 1), the distance of Chebyshev (infinite norm), the distance of Mahalanobis, the distance of Hamming.

It may be desirable to use a weighted distance measurement to determine a change between consecutive LSF vectors. For example, distance d can be calculated according to an expression such as the following:

P)

2

d = ∑ci (li-li),

i = 1

)

in which l indicates the current LSF vector, l indicates the previous LSF vector, P indicates the number of

5 elements in each LSF vector, the index i indicates the vector element LSF, and c indicates a vector of weighting factors. The values of c can be selected to highlight the less frequent components that are significantly more significant. In one example, ci has the value of 1.0 for i between 1 and 8, 0.8 for i = 9, and 0.4 for i = 10. In another example, the distance d between vectors can be calculated Consecutive LSF according to a

10 expression such as the following:

P)

2

d = cw (l - l),

∑ iii i i = 1

In which w indicates a vector of variable weighting factors. In such an example, wi has the value P (fi) r, in which P denotes the LPC power spectrum evaluated at the corresponding frequency f, and r is a constant that has a typical value, for example, of 0.15 or 0.3. In

In another example, the values of w are selected according to a weighted function used in the ITU-T G.729 standard:

1.0 if (2 (l −l) −1)> 0

⎧ π

i + 1 i − 1

w =,

i ⎨ π 2

10 (2 (l −l) −1) +1 otherwise

⎩ i + 1 i − 1

The limit values close to 0 and 0.5 are selected instead of li-1 and li + 1 for the lowest and highest elements of w, respectively. In such cases, ci can have the 20 values indicated above. In another example, ci has the value of 1.0, except for c4 and c5 that

They have the value of 1.2.

It can be seen in FIGURES 4a-d that, frame by frame, a temporary noise formation procedure, as described herein, can increase the quantization error. However, although the absolute square error of the operation of

Quantification can increase, a potential advantage is that the quantization error can be moved to a different part of the spectrum. For example, the quantization error can be moved to the lower frequencies, thus becoming more flat. Since the input signal is also flat, a flatter output signal can be obtained as a sum of the input signal and the flattened quantization error.

FIGURE 7b shows an example of a basic source-filter arrangement as applied to the coding of the spectral envelope of a narrow band signal S20. An analysis module calculates a set of parameters that characterize a filter corresponding to the voice sound over a period of time (usually 20 msec). A bleach filter (also called an analysis or prediction error filter) configured according to those filter parameters that eliminate the spectral envelope to spectrally flatten the signal. The resulting bleached signal (also called residual) has less energy and, therefore, less variation and is easier to encode than the original voice signal. The errors resulting from the coding of the residual signal by the spectrum can also be spread more evenly. Normally, the filter and residual parameters are quantified for efficient transmission through the channel. In the decoder, a synthesis filter configured according to the filter parameters is excited by means of a signal based on the residual signal to produce a synthesized version of the original voice sound. Normally, the synthesis filter is configured to have a transfer function that is the inverse of the transfer function of the bleach filter. FIGURE 8 shows a block diagram of a basic implementation A122 of a narrow band A120 encoder as shown in FIGURE 10a.

As shown in FIGURE 8, a narrowband encoder A122 also generates a residual signal by passing the narrowband signal S20 through a bleach filter 260 (also called an analysis or prediction error filter) that is configured according to the set of filter coefficients. In this particular example, bleach filter 260 is implemented as an FIR filter, although IIR implementations can also be used. Normally, this residual signal will contain perceptually important information of the speech frame, such as a long duration structure related to the fundamental frequency, which is not represented in the S40 parameters of the narrowband filter. The quantizer 270 is configured to calculate a quantized representation of this residual signal to output it as the encoded signal S50 of narrow band excitation. Typically, such a quantifier includes a vector quantifier that encodes the input vector as an index to a corresponding vector entry in a table or code list. Alternatively, such a quantifier may be configured to send one or more parameters from which the vector can be generated dynamically in the decoder, instead of being recovered from storage, as in a poor code procedure. This procedure is used in coding models such as algebraic CELP (linear prediction with code excitation) and in codecs such as 3GPP2 (2nd Third Generation Project) EVRC (Enhanced Variable Transmission Rate Codec).

It is desirable that a narrowband A120 encoder generates the encoded narrowband excitation signal according to the same values of the filter parameters that will be available for the corresponding narrowband decoder. In this way, the encoded signal resulting from narrowband excitation can already explain to some extent the lack of correspondence with the ideal in those parameter values, such as a quantization error. Consequently, it is desirable to configure the bleach filter using the same coefficient values that will be available in the decoder. In the basic example of the encoder A122, as shown in FIGURE 8, the inverse quantizer 240 decrypts the parameters S40 of the narrowband filter, the transform 250 of the filter coefficient from LSF to LP re-establishes a correspondence between the resulting values and a corresponding set of LP filter coefficients, and this set of coefficients is used to configure bleach filter 260 to generate the residual signal that is quantified by quantifier 270.

Some implementations of narrowband A120 encoder are configured to calculate the encoded signal S50 of narrowband excitation to identify one among a set of code vectors that best corresponds to the residual signal. However, it should be noted that the narrowband A120 encoder can also be implemented to calculate a quantified representation of the residual signal without actually generating the residual signal. For example, the narrowband A120 encoder may be configured to use a number of code vectors to generate corresponding synthesized signals (for example, according to a current set of filter parameters), and to select the code vector associated with the signal. generated that best corresponds to the original narrow band signal S20 in a perceptually weighted domain.

FIGURE 9 shows a block diagram of an implementation B 112 of the narrowband decoder B110. An inverse quantizer 310 decrypts the parameters S40 of the narrowband filter (in this case, to a set of LSF), and a transform 320 of the filter coefficient from LSF to LP transforms the LSFs into a set of filter coefficients (for example , as described above with reference to reverse quantifier 240 and transform 250 of narrowband encoder A122). Inverse quantizer 340 decrypts the encoded narrowband excitation signal S50 to produce a narrowband excitation signal S80. Based on the coefficients of the filter and the narrowband excitation signal S80, the narrowband synthesis filter 330 synthesizes the narrowband signal S90. In other words, the narrowband synthesis filter 330 is configured to spectrally shape the narrowband excitation signal S80 according to the quantified coefficients of the filter to produce the narrowband signal S90. As shown in FIGURE 11a, the narrowband decoder B112 (in the form of the narrowband decoder B110) also provides a narrowband excitation signal S80 to the highband decoder B200, which uses it to derive an excitation signal high band In some implementations, the narrowband decoder B110 may be configured to provide additional information to the highband decoder B200 that relates to the narrowband signal, such as spectral division, gain and delay of the fundamental frequency, and Voice mode The A122 narrowband encoder and B112 narrowband decoder system is a basic example of a speech analysis codec by synthesis.

Traditionally, voice communications over the public switched telephone network (PSTN) have been limited in bandwidth to the frequency range of 300-400 kHz. New networks for voice communications, such as cell phone and voice over IP (VoIP), may not have the same bandwidth limits, and it may be desirable to transmit and receive voice communications that include a frequency range of broadband in such networks. For example, it may be desirable to support an audio frequency range that extends up to 50 Hz and / or up to 7 or 8 kHz. It may also be desirable to support other applications, such as high quality audio or the establishment of audio / video conferences, which may have audio voice content at intervals outside the traditional limits of the PSTN.

An approach to broadband voice coding involves scaling a narrowband voice coding technique (for example, one configured to encode the 0-4 kHz range) to cover the broadband spectrum. For example, a voice signal can be sampled at a higher transmission rate to include components at higher frequencies, and a narrowband coding technique can be reconfigured to use more filter coefficients to represent this broadband signal. . However, narrowband coding techniques, such as CELP (linear prediction with code excitation), are computationally intensive, and a broadband CELP encoder can consume too many cycles of procedures to be practical for many applications. mobile and other built-in applications. The coding of the entire spectrum of a broadband signal with a desired quality using said technique can also result in an unacceptably large increase in bandwidth. In addition, a transcoding of said encoded signal would be required before even its narrowband portion could be transmitted in a system, or was decoded by it, that only supports a narrowband coding.

FIGURE 10a shows a block diagram of a broadband voice A100 encoder that includes separate A120 and A200 narrowband and high band voice encoders, respectively. Either or both can be configured to perform the quantification of LSF (or other coefficient representation) using an implementation of quantizer 230, as disclosed herein. FIGURE 11a shows a block diagram of a corresponding broadband voice decoder B100. In FIGURE 10a, the filter bank A110 can be implemented to produce the narrowband signal S20 and the highband signal S30 of a broadband voice signal S10 according to the implementations and principles disclosed in the request for patent

U.S. "SYSTEMS, METHODS, AND APPARATUS FOR SPEECH SIGNAL FILTERING", presented here, which is now U.S. No. 2007/0088558.

It may be desirable to implement a broadband voice coding, so that at least the narrowband portion of the encoded signal can be sent through a narrowband channel (such as a PSTN channel) without a transcoding or Another significant modification. The effectiveness of the broadband coding extension may also be desirable, for example, to avoid a significant reduction in the number of users that can be served in applications such as wireless cellular telephony and broadcast by wired and wireless channels.

An approach to broadband voice coding involves extrapolating the high band spectral envelope from the narrow band coded spectral envelope. However, although such an approach can be implemented without any increase in bandwidth and without a need for transcoding, in general the gross spectral envelope or the structure of the high band portion of a signal cannot be accurately predicted. of voice from the spectral envelope of the narrow band portion.

A particular example of the A100 broadband voice encoder is configured to encode the broadband voice signal S 10 at a transmission rate of approximately 8.55 kbps (kilobits per second), using approximately 7.55 kbps for S40 parameters of the narrowband filter and the encoded signal S50 of narrowband excitation, and approximately 1 kbps being used for the high-band coding parameters S60 (for example, the filter parameters and / or the gain parameters).

It may be desirable to combine the low band and high band encoded signals into a single bit stream. For example, it may be desirable to multiplex the coded signals together to be transmitted (for example, by a wired, optical or wireless transmission channel), or to be stored, as a coded broadband voice signal. FIGURE 10b shows a block diagram of a broadband voice encoder A102 that includes an A130 multiplexer configured to combine parameters S40 of the narrowband filter, an encoded signal S50 of narrowband excitation, and parameters S60 of band coding high on an S70 multiplexed signal. FIGURE 11b shows a block diagram of a corresponding implementation B 102 of the broadband voice decoder B100.

It may be desirable that the A130 multiplexer is configured to embed the low-band coded signal (including parameters S40 of the narrowband filter and the coded signal S50 of narrow-band excitation) as a detachable undercurrent of the multiplexed signal S70, so that the encoded low band signal can be recovered and decoded independently of another portion of the multiplexed signal S70, such as a high band and / or very low band signal. For example, the multiplexed signal S70 may be arranged so that the low band encoded signal can be recovered by eliminating the high band coding parameters S60. A potential advantage of such a feature is that the need to transcode the encoded broadband signal before passing it to a system that supports decoding the low band signal but does not support decoding the high band portion is avoided.

An apparatus that includes a noise formation quantifier and / or a low band, high band and / or broadband voice encoder as described herein may also include circuitry configured to transmit the encoded signal on a channel of transmission such as a wired, optical or wireless channel. Such an apparatus may also be configured to carry out one or more channel coding operations in the signal, such as an error correction coding (for example, a convolutional compatible rate coding) and / or an error detection coding. (for example, a cyclic redundancy coding), and / or one or more layers of the network protocol coding (eg, Ethernet, TCP / IP, cdma2000).

It may be desirable to implement a low band voice A120 encoder as a synthesis analysis speech encoder. Linear Prediction coding with code excitation (CELP) is a popular family of synthesis analysis coding, and implementations of such encoders can perform a coding of the residual signal waveform, including such operations as selection of fixed and adaptable code entries, error minimization operations, and / or weighted perceptual operations. Other implementations of the synthesis analysis coding include the linear prediction coding of mixed excitation (MELP), of algebraic CELP (ACELP), of relaxation CELP (RCELP), of regular pulse excitation (RPE), of CELP of multiple pulses (MPE), and linear prediction with vector sum excitation (VSELP). Related coding procedures include multiband excitation (MBE) and waveform prototype (PWI) interpolation coding. Examples of standardized speech analysis codecs by synthesis include the full rate codec ETSI (European Telecommunications Standards Institute) -GSM (GSM 06.10), which uses a linear prediction excited by residual signal (RELP); the enhanced full rate GSM codec (ETSI-GSM 06.60); the standard ITU (International Telecommunications Union) 11.8 kb / s G.729 Annex E encoder; IS codecs (Interim Standard) 641 for IS-136 (a time division multiple access system); GSM adaptable multitasking (GSM-AMR) codecs; and the 4GV ™ codec (Fourth Generation Vocoder ™) (QUALCOMM Incorporated, San Diego, California, USA). Existing implementations of the RCELP encoders include the Enhanced Variable Rate Transmission Codec (EVRC), as described in the Telecommunications Industry Association (TIA) IS127, and the Second Joint Third Generation Project (3GPP2) Vocoder selectable mode (SMV). The various low band, high band, and broadband encoders described herein may be implemented according to any of these technologies, or any other voice coding technology (whether known or still undeveloped) that represents a voice signal such as (A) a set of parameters that describe a filter and (B) a quantified representation of a residual signal that provides at least part of an excitation used to activate the described filter to reproduce the voice signal.

As mentioned above, the embodiments as described herein include implementations that can be used to perform embedded coding, supporting compatibility with narrowband systems and avoiding the need for transcoding. Support for high band coding can also serve to differentiate based on cost between chips, chipsets, devices, and / or networks that have broadband support with a backward compatibility, and those that have only one band support narrow. A support for an encoding as described herein can also be used in conjunction with a technique to support a low band encoding, and a system, method or apparatus according to such an embodiment can support an encoding of frequency components from, for example , about 50 or 100 Hz to about 7 or 8 kHz.

As mentioned above, adding a high band support to a voice encoder can improve intelligibility, especially with regard to the differentiation of fricatives. Although a human listener can usually derive such differentiation from the particular context, a high band support can serve as an enabling feature in speech recognition and other machine interpretation applications, such as systems for automated navigation of menus by voice and / or automatic call processing.

An apparatus according to one embodiment may be incorporated in a portable device for wireless communications, such as a cell phone or a personal address book (PDA). Alternatively, such an apparatus may be included in another communications device, such as in a VoIP handset, in a personal computer configured to support VoIP communications, or in a network device configured to route telephone or VoIP communications. For example, an apparatus can be implemented according to an embodiment on a chip or chipset for a communication device. Depending on the particular application, said device may also include such features as an analog to digital and / or digital to analog conversion of a voice signal, circuitry to carry out the amplification and / or other signal processing operations in a voice signal, and / or radiofrequency circuitry for the transmission and / or reception of the encoded voice signal.

It is contemplated and explicitly disclosed that the embodiments may include and / or be used with any one or more of the other features disclosed in the provisional U.S. patent application. No. 60 / 667,901, now U.S. No. 2007/0088542. Such features include the displacement of the high-band signal S30 and / or the high-band excitation signal S120 according to a regularization or other displacement of the narrow-band excitation signal S80 or the narrow-band residual signal S50. Such features include an adaptive flattening of LSFs, which can be carried out before quantification as described herein. Such features also include a fixed or adaptable flattening of a gain of the envelope, and an adaptable attenuation of a gain of the envelope.

The above presentation of the described embodiments is provided to allow any person skilled in the art to make or use the present invention. Various modifications to these embodiments are possible, and the generic principles presented herein can also be applied to other embodiments. For example, an embodiment may be implemented in part or in whole as a wired circuit, as a circuit configuration manufactured in a specific application integrated circuit, or as an unalterable software loaded in a non-volatile storage or a software program. software loaded with or in a data storage medium as machine-readable code, said code being instructions that can be executed by a set of logical elements such as a microprocessor or other digital signal processing unit. The data storage medium may be a set of storage elements such as semiconductor memory (which may include without limitation dynamic or static RAM (random access memory), ROM (read-only memory), and / or flash RAM ), or ferroelectric, magnetoresistive, ovonic, polymeric, or phase change memory; or a disk medium such as a magnetic or optical disk. It should be understood that the term "software" includes source code, assembly language code, machine code, binary code, firmware, macrocode, microcode, any one or more sets or sequences of instructions that can be executed by a set of logical elements, and any combination of such examples.

The various elements of implementations of a noise formation quantifier can be implemented; of an A200 high band voice encoder; of an A100 and A102 broadband voice encoder; and arrangements that include one or more devices such as electronic and / or optical devices that reside, for example, on the same chip or between two or more chips in a chipset, although other provisions are also contemplated without such limitation. One or more elements of such apparatus can be implemented in whole or in part as one or more sets of instructions arranged to execute in one or more fixed or programmable sets of logical elements (eg, transistors, gates) such as microprocessors, built-in processors , IP cores, digital signal processors, FPGA (programmable field gate matrices), ASSP (specific application standard products) and ASIC (specific application integrated circuits). It is also possible that one or more such elements have a common structure (for example, a processor used to execute portions of code corresponding to different elements at different times, a set of instructions executed to carry out tasks corresponding to different elements in different moments, or an arrangement of electronic and / or optical devices that carry out operations for different elements at different times). In addition, one or more of such elements may be used to perform tasks or execute other sets of instructions that are not directly related to an operation of the device, such as a task related to another operation of a device or system in the device. that the device is incorporated.

The embodiments also include additional procedures for voice processing, voice coding, and suppression of high-band bursts as expressly disclosed herein, for example, by means of descriptions of structural embodiments configured to carry carry out such procedures. Each of these procedures can also be tangibly implemented (for example, in one or more data storage media as listed above) as one or more sets of instructions readable and / or executable by a machine, including a set of elements. logic (for example, a processor, a microprocessor, a microcontroller, or other finite state machine). Therefore, it is not intended that the present invention be limited to the embodiments shown above but, rather, should be granted the broadest scope defined by the appended claims.

Claims (32)

one.

A procedure for signal processing, said procedure comprising:

encoding a first frame and a second frame of a voice signal to produce corresponding first and second vectors, in which the first vector represents a spectral envelope of the voice signal during the first frame and the second vector represents a spectral envelope of the voice signal during the second frame; generating a first quantified vector, including said generation quantifying a third vector that is based on the first vector, calculating a quantization error of the first quantized vector; calculate a fourth vector, said calculation including adding a scale version of the quantization error to the second vector; and quantify the fourth vector.

2.

The method according to claim 1, wherein said calculation of a quantization error includes calculating a difference between the first quantized vector and the third vector.

3.

The method according to claim 1, wherein said calculation of a quantization error includes calculating a difference between the first quantized vector and at least a portion of the first vector.

Four.

The method according to claim 1, said method including calculating the quantification error at scale, said calculation comprising multiplying the quantization error by a scale factor, in which the scale factor is based on a distance between at least a portion of the first vector and a corresponding portion of the second vector.

5.

The method according to claim 4, wherein each of the first and second vectors includes a plurality of line spectral frequencies.

6.

The method according to claim 1, wherein each of the vectors

First and second includes a representation of a plurality of linear prediction filter coefficients.

7.

The method according to claim 1, wherein each of the first and second vectors includes a plurality of line spectral frequencies.

8.

An apparatus comprising:

means for encoding a first frame and a second frame of a voice signal to produce corresponding first and second vectors, in which the first vector represents a spectral envelope of the voice signal during the first frame and the second vector represents an envelope spectral of the voice signal during the second frame; means for generating a first quantified vector, said generation including quantifying a third vector that is based on the first vector; a means to calculate a quantization error of the first quantized vector; and a means for calculating a fourth vector, including said calculation adding a scale version of the quantization error to the second vector, wherein said means for generating a first quantified vector is configured to quantify the fourth vector.

9.

The apparatus according to claim 8, wherein:

said means for encoding comprises a voice encoder; said means for generating comprises a quantifier; said means for calculating a quantization error of the first quantized vector it comprises a first adder; Y said means for calculating a fourth vector comprises a second adder.

10.

The apparatus according to claim 9, wherein said first adder is configured to calculate the quantization error based on the difference between the first quantized vector and the third vector.

eleven.

The apparatus according to claim 9, wherein said first adder is configured to calculate the quantization error based on a difference between the first vector

quantified and at least a portion of the first vector.

12.

The apparatus according to claim 9, said apparatus including a multiplier configured to calculate the quantification error at scale based on a product of the quantization error and a scale factor, wherein said apparatus includes logic configured to calculate the scale factor based on a distance between at least a portion of the first vector and a corresponding portion of the second vector.

13.

The apparatus according to claim 12, wherein each of the first and second vectors includes a plurality of line spectral frequencies.

14.

The apparatus according to claim 9, wherein each of the first and second vectors includes a representation of a plurality of linear prediction filter coefficients.

fifteen.

The apparatus according to claim 9, wherein each of the first and second vectors includes a plurality of line spectral frequencies.

16.

The apparatus according to claim 9, said apparatus comprising a device for wireless communications.

17.

The apparatus according to claim 9, said apparatus comprising a device configured to transmit a plurality of packets compatible with a version of the Internet Protocol, wherein the plurality of packets describes the first quantized vector.

18.

A computer-readable medium comprising instructions that, after being executed in a processor, causes the processor to carry out the steps of the process of any one of claims 1 to 6.

19.

The method according to claim 1, wherein the second frame immediately follows the first frame in the voice signal.

twenty.

The method according to claim 1, wherein each of the vectors first and

second represents an adaptively flattened spectral envelope.

twenty-one.

The method according to claim 1, wherein said method comprises:

to quantify the fourth vector; Y calculate an excitation signal based on the fourth unquantified vector.

22

The method according to claim 1, wherein said method comprises filtering a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, and wherein the first vector represents a spectral envelope of the narrowband voice signal during the first frame, and in which the second vector represents a spectral envelope of the narrow band voice signal during the second frame.

2. 3.

The method according to claim 1, wherein said method comprises filtering a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, and wherein the first vector represents a spectral envelope of the high band voice signal during the first frame, and in which the second vector represents a spectral envelope of the high band voice signal during the second frame.

24.

The method according to claim 1, wherein said method comprises:

filter a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, in which (A) the first vector represents a spectral envelope of the narrow band voice signal during the first frame and (B) the second vector represents a spectral envelope of the narrowband voice signal during the second frame; to quantify the fourth vector; based on the fourth unquantified vector, calculate an excitation signal for the narrowband voice signal; and, based on the excitation signal for the narrowband voice signal, derive an excitation signal for the high band voice signal.

25.

The method according to claim 1, wherein said fourth quantification

vector comprises carrying out a quantification of the separator vector of the fourth vector.

26.

The apparatus according to claim 8 or 9, wherein the second frame immediately follows the first frame in the voice signal.

27.

The apparatus according to claim 8 or 9, wherein each of the first and second vectors represents an adaptively flattened spectral envelope.

28.

The apparatus according to claim 8 or 9, wherein said apparatus comprises:

a means to quantify the fourth vector; and a means to calculate an excitation signal based on the fourth unquantified vector.

29.

The apparatus according to claim 8 or 9, wherein said apparatus comprises means for filtering a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, and wherein the first vector represents a spectral envelope of the narrowband voice signal during the first frame, and in which the second vector represents a spectral envelope of the narrow band voice signal during the second frame.

30

The apparatus according to claim 8 or 9, wherein said apparatus comprises means for filtering a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, and wherein the first vector represents a spectral envelope of the high band voice signal during the first frame, and in which the second vector represents a spectral envelope of the high band voice signal during the second frame.

31.

The apparatus according to claim 8 or 9, wherein said apparatus comprises:

means for filtering a broadband voice signal to obtain a narrow band voice signal and a high band voice signal, in which (A) the first vector represents a spectral envelope of the narrow band voice signal during the first frame and (B) the second vector represents an envelope spectral of the narrowband voice signal during the second frame; a means to quantify the fourth vector; a means to calculate an excitation signal for the band voice signal narrow based on the fourth unquantified vector; Y 5 means for deriving an excitation signal for the high band voice signal based on the excitation signal for the narrow band voice signal. 32. The apparatus according to claim 8 or 9, wherein said means for generating a first Quantified vector is configured to quantify the fourth vector by performing a quantification of the separator vector of the fourth vector.