JP2003195899A - Encoding method for speech/sound signal and electronic device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an encoding method for speech/sound signals which can generate the speech signal and sound signal even at a low bit rate. <P>SOLUTION: The encoding method for the speech/sound signals which includes encoding of pulse position information comprises a step for obtaining a parameter representing a short-time spectrum of an input signal, a step for finding an impulse response on the basis of the obtained parameter representing the short-time spectrum, a step for obtaining at least position weight information from the input signal, and a step for selecting the pulse position information by using the impulse response and position weight information. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio / audio signal encoding method and an electronic device.

[0002]

2. Description of the Related Art C is a method for compressing and encoding a voice signal.
An ELP (Code-Excited Linear Prediction) method is known. For the CELP method, for example, ["Code-
Excited Linear Prediction (CELP): High-quality Speec
h at Very Low Rates "Proc.ICASSP '85,25,1.1.pp.937
-940, 1985].

In the CELP system, a voice signal is modeled by being divided into a synthesis filter and a sound source signal for driving the synthesis filter. The encoded synthesized speech signal is generated by passing the excitation signal through the synthesis filter.

The excitation signal is generated by combining two code vectors, that is, an adaptive code vector generated from an adaptive codebook storing past excitation signals and a noise vector generated from a noise codebook.

The adaptive code vector mainly has a role of representing the repetition of the waveform due to the pitch period, which is a feature of the sound source signal in the voiced sound section.

On the other hand, the noise code vector has a role of supplementing the components contained in the excitation signal that cannot be represented by the adaptive code vector, and is used to make the synthesized speech signal more natural. Only the adaptive code vector or the noise code vector may be referred to as the excitation signal.

[0007] The CELP method is characterized in that the excitation signal is encoded by evaluating the distortion at the level of the auditory-weighted audio signal so that the encoded distortion is less perceptible.

FIG. 12 shows a process of generating a perceptually weighted combined voice signal from a sound source signal by perceptually weighted synthesis. The candidate of the residual level sound source signal from the sound source signal generation unit 300 is input to the perceptual weighting synthesis unit 301 via the path 305, and the perceptual weighted synthetic speech signal is output to the output terminal 304.

The reason why the coding distortion is hardly perceived by using the perceptual weight is that the perceptual weighting is performed so that the spectrum of the coding distortion is masked in the shape of the spectrum of the voice signal. is doing. The perceptual weighting is obtained from the audio signal for each coding section, and the same perceptual weighting characteristic is used in the same coding section.

As described above, the conventional coding is characterized in that the perceptual weighting is obtained from the speech signal for each coding section, and the excitation signal is coded using the same weighting characteristic in the coding section.

In such a conventional method, when the coding bit rate is lowered to about 4 kbit / s in the case of a voice signal, for example, the number of bits allocated for expressing the excitation signal becomes insufficient, so that the distortion due to the coding is generated. Will be perceived as sound. As a result, the deterioration of the sound quality such as the faint sound and the mixing of noise becomes remarkable.

Therefore, there is a demand for highly efficient coding that can generate high-quality synthesized speech even if the bit rate is reduced. Such a requirement also applies to the encoding of acoustic signals.

[0013]

As described above, in the conventional speech / audio signal coding method, the perceptual weighting is obtained from the speech signal for each coding section, and the same weighting characteristic is used in the coding section. Since the sound source signal is encoded,
There is a problem in that it is difficult to obtain high quality synthetic speech at a low bit rate.

The present invention has been made in view of these problems, and an object thereof is to provide a voice / audio signal capable of generating a high-quality voice signal / acoustic signal even at a low bit rate.
An object is to provide an audio signal encoding method and an electronic device.

[0015]

In order to achieve the above object, the first invention is a method of encoding a voice / acoustic signal including encoding of pulse position information, which is a short-time spectrum of an input signal. , A step of obtaining an impulse response based on the acquired parameter representing the short-time spectrum, a step of obtaining at least position weight information from an input signal, the impulse response and the position weight information. Using the pulse position information.

A second aspect of the present invention is a method for encoding a voice / acoustic signal including encoding of pulse position information, which comprises a step of acquiring a parameter representing a short-time spectrum of an input signal and the acquired short-time. Obtaining an impulse response based on a parameter representing a spectrum, obtaining at least position weight information from an input signal, calculating a position weighted correlation value using the impulse response and the position weight information, and obtaining And selecting the pulse position information using the position-weighted correlation value.

A third aspect of the present invention is a method for encoding a voice / acoustic signal including encoding of pulse position information, which comprises a step of acquiring a parameter representing a short-time spectrum of an input signal and the acquired short-time. A step of obtaining an impulse response based on a parameter representing a spectrum, a step of obtaining at least a target signal and position weight information from an input signal, and a step of obtaining a position weighted correlation value based on the impulse response and the position weight information. A step of obtaining a position-weighted cross-correlation value using the impulse response, the target signal, and the position-weight information, and selection of the pulse position information using the position-weighted correlation value and the position-weighted cross-correlation value And a step of performing.

A fourth invention uses the algebraic codebook for coding the pulse position information in the audio / audio signal coding method according to any one of the first to third inventions.

A fifth aspect of the invention is the method of encoding a voice / acoustic signal according to any one of the first to fourth aspects, further comprising a step of obtaining a short-term prediction residual signal from the input signal. The position weight information is obtained using the short-term prediction residual signal.

A sixth aspect of the present invention is a voice / acoustic signal encoding method including encoding of information representing a sound source signal,
The step of acquiring a parameter representing the short-time spectrum of the input signal, the step of obtaining an impulse response based on the acquired parameter representing the short-time spectrum, and the position weight information using the short-term prediction residual signal acquired from the input signal. The method further comprises a step of obtaining and a step of selecting information representing the sound source signal using the impulse response and the position weight information.

A seventh aspect of the invention uses the weighting of both the position weighting and the perceptual weighting in the method of encoding a voice / acoustic signal according to any one of the first to sixth aspects of the invention.

An eighth aspect of the present invention is a voice / acoustic signal encoding method including encoding of information representing a sound source signal,
The method further comprises the steps of position-weighting the candidate signal for representing the sound source signal, and selecting the candidate signal based on the distortion when the position-weighted candidate signals are combined.

A ninth aspect of the present invention is a voice / acoustic signal encoding method including encoding of information representing a sound source signal,
Positioning is performed on the candidate signal for representing the sound source signal, and selection of the candidate signal is performed based on distortion when the position-weighted candidate signal is auditorily weighted combined.

A tenth aspect of the invention is an input section for inputting a voice / acoustic signal, and an encoding section for performing an encoding process on the voice / acoustic signal input via the input section. A transmitter for transmitting the voice / acoustic signal encoded by the encoder, a receiver for receiving the encoded voice / acoustic signal, and a voice / acoustic signal received via the receiver. And a decoding unit for performing a decoding process, and an output unit for outputting the voice / sound signal decoded by the decoding unit, wherein the coding unit is any one of the first to ninth inventions.
The encoding method described in No. 1 is executed.

[0025]

DETAILED DESCRIPTION OF THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 shows the principle of residual level position weighting in the encoding method of the present invention.
In the figure, the sound source signal candidates of the residual level from the sound source signal generation unit 300 are input to the residual level position weighting unit 302 via a path 308. Here, the position-weighted residual sound source signal candidates are
09, and is input to the perceptual weighting synthesis unit 303, and the perceptual weighted combined voice signal with residual level position weighting is output to the output terminal 310.

By selecting the code representing the sound source signal by using such a residual level position-weighted auditory weighted synthetic speech signal, the information of the positionally important signal at the residual signal level can be more accurately measured. Since it becomes possible to incorporate the masking effect by weighting of hearing while reflecting the coding, the performance of the coding is improved.

FIG. 2 shows the principle configuration of the coding method of the present invention for performing code selection of the random codebook by using the position weighting of the residual signal level and the distortion evaluation at the perceptually weighted synthesis level. , The main part of the encoding of the excitation signal used to represent the input signal. The target residual signal generation unit 901 includes a spectrum parameter processing unit 900.
The target residual signal is generated by using the parameter representing the short-time spectrum of the input signal acquired in (1). The target residual signal corresponds to the ideal signal of the sound source signal.

On the other hand, the random codebook 902 outputs code vector candidates. The residual level position weight acquisition unit 903 acquires position weight information suitable for position weighting of the residual level using the input signal and the parameter representing the short time spectrum. The residual level position weighting section 904 and the residual level position weighting section 905 have a function of performing position weighting on the residual level signals of the target residual signal and the code vector candidate, respectively.

The perceptually weighted synthesis unit 906 outputs the position-weighted target residual signal to the distortion evaluation unit 909 as a perceptually weighted target signal with position weighting. Similarly, the auditory weighted synthesis unit 907 converts the position weighted code vector candidate into a position weighted auditory weighted synthetic code vector candidate, further multiplies the gain by the gain multiplication unit 908, and outputs the result to the distortion evaluation unit 909. Here, the distortion evaluation value of the code candidate is obtained for the target. This operation is performed for each predetermined code candidate in the random codebook 902, the code selection unit 910 selects a code candidate having a more favorable distortion evaluation value, and the code C is output to output the code of the excitation signal. Is realized.

FIG. 3 is a block diagram showing an example of a coding unit which realizes the method of coding a voice / acoustic signal of the present invention. Here, an example in which the present invention is applied to CELP coding of a voice signal will be described.

The input voice input from a voice input means (not shown) such as a microphone is subjected to A / D conversion, and is input to the spectrum parameter analysis / encoding unit 500 of FIG. 3 for each predetermined time interval. . Usually this time section is 10-30m
A length of about s is used and is sometimes called a frame.

In the CELP method, as a model of a voice generation process, a vocal cord signal is made to correspond to a sound source signal, a spectral envelope characteristic represented by the vocal tract is expressed by a synthesis filter, the sound source signal is input to the synthesis filter, and the output of the synthesis filter is used. Represents an audio signal. The present invention is the same as the conventional CELP system in the general framework that the code selection of the sound source signal is performed so that the waveform distortion between the input speech signal and the synthesized speech signal becomes small.

However, in the present invention, (1) the point weight is introduced into the codebook search to evaluate the waveform distortion, and (2) the signal of the residual signal level is used instead of the signal of the voice waveform level. The point that the position weighting is applied to a certain sound source signal or sound source pulse is very different from the conventional one. Here, the algebraic codebook (Algebr
Although description will be made assuming the use of an aic codebook), the essence of the present invention is only in (1) and (2) above, and the structure of the codebook itself is not limited to the algebraic codebook representing excitation pulses. For example, the present invention is effective even if the codebook is a codebook for representing multiple pulses. In the present invention, these codebooks will be collectively referred to as a random codebook.

Spectral parameter analysis / encoding unit 50
0 inputs a voice signal for each frame and analyzes the input voice signal to extract spectrum parameters. Then, using the extracted spectral parameters,
A spectrum parameter codebook prepared in advance is searched for, and an index (spectrum parameter code) A of the codebook capable of better expressing the spectrum envelope of the input signal is selected and output. Usually CE
In the LP method, LSP (Line Spectrum Pai) is used as a spectrum parameter used when encoding a spectrum envelope.
r) parameter is used, but it is not limited to this, and LP can be used as long as it can express the spectrum envelope.
Other parameters such as the C factor are also valid.

The target signal calculation unit 511 receives the input voice signal, the spectrum parameters from the spectrum parameter analysis / encoding unit 500, and the excitation signal generation unit 51 described later.
The excitation signal from 0 is used to calculate the target signal X (n) for encoding in adaptive codebook search section 504.

The impulse response calculation section 501 obtains an impulse response h (n) based on the spectrum parameter from the spectrum parameter analysis / encoding section 500. This impulse response is typically calculated using a perceptual weighted synthesis filter H (z) that is a combination of a synthesis filter using LPC coefficients and a perceptual weighting filter.

[0038]

[Equation 1]

Residual signal calculating section 502 calculates a residual signal using the input speech and the spectrum parameter from spectrum parameter analyzing / encoding section 500. As a specific example, an LPC coefficient is used as the extracted spectrum parameter, and a prediction residual signal is obtained by filtering the audio signal with the prediction filter A (z) using the LPC coefficient. The prediction residual signal r (n) can be obtained using the audio signal s (n), for example, as follows.

[0040]

[Equation 2]

Here, α _i is an LPC coefficient which is not quantized, but a prediction residual signal similar to r (n) can be obtained by substituting the quantized LPC coefficient. Formula (2) is LP
The residual signal obtained by this method is called a short-term predicted residual signal because it is a prediction using data of P samples in the vicinity using the C coefficient. The prediction residual signal may also be called a residual signal or simply a residual. In the following description, it will be referred to as a residual signal or a residual.

The position weight setting unit 503 is a residual signal calculation unit 5
A position weight is obtained based on the residual signal obtained in 02 and output to the position weighted correlation calculation unit 505 and the position weighted cross correlation calculation unit 506, respectively. In this embodiment,
For simplification of the description, an example in which the position weight is used only for the code search of the algebraic codebook will be described, but the adaptive codebook search unit 504
It goes without saying that the position weight can be applied to the gain codebook search unit 509. In that case, the position weight is also output to the adaptive codebook searching unit 504 and the gain codebook searching unit 509.

FIGS. 4A to 4C and FIG. 5A,
(B) is a diagram for explaining an example of a method for obtaining the position weight in the position weight setting unit 503. For simplicity of explanation, the frame length is 24 samples (= position is 2
However, it is needless to say that the present invention includes the case where a frame is divided into a plurality of subframes and the position weight or the code search of the codebook is performed for each subframe.

FIG. 4A shows a speech signal s (n) before encoding.
Is an example of a discrete waveform of. In the figure, the waveform amplitude of the audio signal at the position n = i is represented by s (i). FIG. 4B shows FIG.
It is a waveform example of the residual signal r (n) obtained from the audio signal of (A). Since the residual signal is an error signal when the speech signal is predicted, it can be said that a position where the amplitude of the residual signal is large is a position that cannot be sufficiently expressed by the prediction. It is considered that the residual signal at that position includes more voice features that cannot be expressed by prediction than other positions with small amplitude.

Therefore, by introducing into the encoding of the excitation signal a mechanism for encoding a position where the amplitude of the residual signal is large with higher accuracy (that is, less distortion) than other positions, a higher quality synthesized speech is provided. It becomes possible to do.

The present invention analyzes at which position the distortion should be made smaller by capturing its characteristics based on the residual signal, and the penalty of the distortion evaluation becomes large at such a position. Set the position weight relatively large.

A specific example for setting the position weight v (n) of each position n according to the relative magnitude relation of each amplitude value r (n) of the residual signal is as follows.

The threshold value TH is calculated from r (n), and if | r (n) |> TH, v (n) = k1 | r (n) | ≤TH, v (n) = k2 where , K1, k2 have a relationship of k1>k2> 0, a large position weight k1 is set at a position where the absolute value amplitude is large. Setting k1 = k2 is equivalent to not using position weight.

An example of a method for setting the position weight from the residual signal will be described below with reference to FIG. 4 (C).

In the figure, the simplest method of comparing the absolute value amplitude at each position of the residual signal with the threshold level 49 determined by a predetermined method and setting the position weight based on the magnitude relationship is shown. There is. That is, if the absolute value amplitude of the residual signal at each position is smaller than the threshold value 49, the position weight is set relatively small, and conversely, the absolute value amplitude is set to the threshold value 49.
If it is larger than that, the position weight is set relatively large.

In the example of FIG. 4C, in fact, the absolute value amplitude indicated by 50 is smaller than the threshold value 49, so the position weight at this position is set relatively small, and the absolute value amplitude indicated by 51 is Since it is larger than the threshold value 49, the position weight at this position is set relatively large.

Although the threshold value TH is one type,
It is also effective to set the position weight value more finely by using multiple thresholds such as 1 and TH2. The threshold value can be determined based on, for example, the square root of the sum of squares of the residual signals, the average of the absolute values, or the variance. If the amplitude of the residual signal is normalized, it is possible to set the position weight with the threshold value set to a substantially fixed value.

The position weight v obtained as a result is shown in FIG.
An example of (n) is shown. In this example, the position weight v (n) has two kinds of values, large (k1) and small (k2). Also,
As can be seen from the figure, all the position weights v (n) of the present invention have the same polarity (all positive: v (n)> 0 in the figure). This indicates that the position weight is a weight function associated with the sample position n.

Since the sampling position n indicates the position n of the sampled time series signal, the position n in the present invention may be considered as the time n or the time n. Therefore, it can be said that the weight v (n) relating to the position is also the position weight relating to the sample position in the target coding section, and is also the time weight (or time weight) relating to the time n defined in this section. You can say that. The weighting relating to such a time position is a weighting defined to be multiplied for each individual sample of the time series signal,
The weighting is completely different from the weighting realized by the filter calculation and the convolution calculation used in the conventional auditory weighting.

FIG. 5B is an example in which the method of setting the position weight at a position where the absolute value of the residual difference is very small is set to a small value and the size of the position weight is set to three types.
For example, in FIG. 4C, the position weight v (21) in FIG. 4 is smaller than the value v (21) in FIG.
This is because the fact that the absolute value amplitude of the residual at position n = 21 is very small is reflected.

As another method of setting the position weight, the residual signal r (n) or a signal obtained by normalizing the residual signal is used,
It is also effective to use a quantized absolute value as the position weight v (n). That is, v (n) = Q [abs (r
(N))]. Here, abs () represents a function representing an absolute value, and Q [x] represents a quantized output when x is input to a predetermined quantum device Q. When a quantizer having a binary quantization output is used, position weights of two different sizes can be set as in the case of FIG. 5A.

Similarly, when a quantizer whose quantized output is ternary is used, three kinds of position weights can be set as in the case of FIG. 5B. There may be four or more types of position weight magnitudes. Also, v
(N) = abs (Q [r (n)]) may be used. If absolute value processing is incorporated in the quantizer, simply v (n)
= Q [r (n)].

As another method of setting the position weight,
Squared signal of residual signal instead of r (n) {r (n)} ²
It is also possible to set the position weight by using the method described in the above example.

The present invention also includes a method of using a simulated signal having a shape relatively close to the residual signal instead of the residual signal. As such a simulated signal of the residual signal, for example, the adaptive code vector described below can be considered, and it is also effective to use the adaptive code vector instead of the residual signal to obtain the position weight.

As described above, various methods can be considered for setting the position weight, but the point is that the position weight should have a mechanism capable of reflecting the importance of each position on the position weight. The method of determination is also included in the present invention.

Above, FIGS. 4A to 4C and FIG.
The description of (A) and (B) ends.

Now, returning to FIG. 3, the description will be continued.

Adaptive codebook search section 504 is used to represent a component repeated in the pitch period included in the excitation signal. In the CELP method, the coded past excitation signal is stored as an adaptive codebook for a predetermined length, and the adaptive codebook is stored in both the speech coding unit and the speech decoding unit, so that the specified pitch period is supported. The structure is such that the repeated signal can be extracted from the adaptive codebook.

In the adaptive codebook, the output signal from the codebook and the pitch period have a one-to-one correspondence, so the pitch period can correspond to the index of the adaptive codebook. Under such a structure, adaptive codebook searching section 504 searches for a pitch period such that distortion of the synthesized signal and the target signal when the output signal from the adaptive codebook is synthesized by the synthesis filter becomes small. Then, the index (adaptive code) L corresponding to the searched pitch period is output.

The adaptive codebook search unit 504 synthesizes the output signal (adaptive code vector) from the adaptive codebook obtained at the pitch period corresponding to the index (adaptive code) L and the adaptive code vector with the spectrum parameter or impulse response. The combined adaptive code vector obtained by Further, by subtracting the contribution from the adaptive code vector from the target signal X (n), the target signal X2 (n) (hereinafter also referred to as the target vector X2) used in the next noise codebook search is output.

The position weighted correlation calculation unit 505 calculates the position weighted correlation using the position weight and the impulse response h (n). The position-weighted cross-correlation calculation unit 506 calculates the position-weighted cross-correlation using the position weight, the impulse response h (n), and the target signal X2 (n).

Next, the noise codebook is searched using the calculated position-weighted correlation and the position-weighted cross-correlation. Before that, the position-weighted correlation and cross-correlation were used. The principle of the codebook search method will be described below.

The error vector E by the codebook search when the position weight is introduced is expressed as follows.

[0069]

[Equation 3]

Here, H is an impulse response matrix composed of impulse responses h (n), V is a position weight matrix, ck is a code vector corresponding to the code k output from the code book,
g is the gain, r2 is the target residual vector, and X2 =
There is a relationship of Hr2. In the position weight matrix V, in the present embodiment, the main diagonal component vii is the position weight v (i), and the other elements have a value of zero.

[0071]

[Equation 4]

X3 is a position weighted target vector. The impulse response matrix H represents a convolution operation by the auditory weighted synthesis filter H (z) in the form of a matrix, and is a lower triangular matrix having h (0) in the main diagonal component and a matrix with the same diagonal elements. Is.

[0073]

[Equation 5]

Therefore, the error E defined by the equation (3) is
It can be seen that there is an error between the position-weighted target vector X3 and the vector gHVck obtained by synthesizing the position-weighted vector Vck of the code vector ck with the auditory-weighted synthesizing filter H (z) and multiplying it by the gain g. Vr2 and V
ck can be called a position-weighted target residual vector and a position-weighted code vector, respectively.

Therefore, HVr2 (= X3) is a position-weighted target vector, and HVcc is a position-weighted composite code vector. In addition, position weighting is applied to the residual level vectors such as the code vector ck and the residual vector r2 in the equation (1), and the error E is defined at the level where the auditory weighting synthesis is performed on the vector. Please note that.

As mentioned above, r2 is the target residual vector, which is obtained by removing the influence of convolutional synthesis by the impulse response from the target vector X2, and there is a relation of X2 = Hr2. Therefore, r2 can be calculated as r2 = H ⁻¹ X2 by using the target vector X2 and the impulse response h (n). This calculation changes the expression,
The same as below.

[0077]

[Equation 6] Here, L represents the number of dimensions of the vector.

Since the target residual vector r2 (n) is a kind of short-term predicted residual signal generated from the target signal (target vector) X (n), it also contains information on the important position of the sound source signal. It is thought that Therefore, it should be added here that the position weight information can be extracted even if the target residual vector r2 (n) is used instead of the residual signal r (n).

In the search of the algebraic codebook performed by the conventional method, the correlation obtained from the impulse response is the cross-correlation between the target signal and the impulse response (X2 ^t H when expressed as a matrix).
, T represents the transpose of the matrix), and the autocorrelation of the impulse response (corresponding to H ^t H when expressed in a matrix) is required. An algorithm for searching an algebraic codebook at high speed using these correlations is an algebraic codebook (Algebr
Widely known as aic Codebook search method, PDC-EF of mobile phone used in Japan
R standard ACELP system and ITU-T standard voice coding system G.I. 729, and also various known speech coding standards such as AMR which is a speech coding method of the 3GPP standard.

On the other hand, the search of the algebraic codebook using the position weight according to the present invention is as follows.

Now, when the gain g that minimizes the error power E ^t E of the error vector E of the equation (3) is obtained,

[Equation 7] Then, by substituting this into equation (3), the minimum error power value (E ^t E) min is

[Equation 8] Becomes Search codebook is to explore the code k code vectors which maximize the equation (8) on the right side evaluation value of the second term ^{(X3 t HVck) 2 / (} ck t V t H t HVck) ( right side The first term is a fixed value that does not depend on the code k and can be ignored).

On the other hand, in the conventional codebook search using no position weight, the code k of the code vector that maximizes the evaluation value (X2 ^t Hc ^k ) ² / (c ^{k t} H ^t Hc k) is searched. In the conventional method, the cross-correlation X2 ^t H and the auto-correlation H ^t H are obtained in advance before the search so that a suitable code k that minimizes the error power can be searched at high speed by using these values. be able to. However, a method for rapidly searching for a suitable code k that minimizes the error power under the condition of introducing the position weight has not been considered so far, and the present invention provides this method.

From the comparison between the above-mentioned evaluation values obtained by the conventional method and the method of the present invention, the position-weighted cross-correlation X3 ^t HV (= r
2 ^t V ^t H ^t HV) and position-weighted autocorrelation V ^t H ^t H
It is understood that if V is obtained in advance before the codebook search, the optimum code can be selected at high speed by using the conventional algebraic codebook search method. Therefore, if the position-weighted cross-correlation and the position-weighted autocorrelation can be obtained with a small amount of calculation, the amount of calculation required for the codebook search using the position weight of the present invention is similar to that of the conventional method. Can be suppressed to
A coding method more suitable for practical use can be provided.

First, a specific example of the method for obtaining the position-weighted autocorrelation according to the present invention will be described. The element φ (i, i, j) of the autocorrelation H ^t H of the conventional impulse response
j) is

[Equation 9] And H ^t H becomes a symmetric matrix, so φ
There is a relationship of (j, i) = φ (i, j). On the other hand, the method of the present invention requires the calculation of the position-weighted autocorrelation V ^t H ^t HV. However, when the position weight matrix V is a diagonal matrix defined by the equation (4), V ^t H ^t HV Element at row i and column j of
(I, j) can be calculated as follows.

[0085]

[Equation 10]

Here, v (i) is a position weight and is always given as a positive value (that is, v (i)> 0).

When V is a diagonal matrix, the position-weighted autocorrelation V ^t H ^t HV is also a symmetric matrix, so Φ (j, i) =
There is a relationship of Φ (i, j). Therefore, the increase in the amount of calculation for obtaining the position-weighted autocorrelation by the method of the present invention is due to the autocorrelation φ of the impulse response obtained by the conventional method.
It can be seen that it is sufficient to multiply (i, j) by v (i) v (j).

Furthermore, only in a limited number of positions in advance,
It is clear that if the position weight is set so as to use a position weight different from 1, the calculation amount of the equation (10) can be further reduced.

In a typical algebraic codebook search method, the pulse amplitude for each position is set to either +1 or -1 before the search for the excitation pulse position, and the autocorrelation φ '( The calculation amount of the search is reduced by performing the search of the pulse position using i, j). That is, φ ′ (i, j) by the conventional method is

[Equation 11] Can be found at. On the other hand, according to the present invention, position-weighted autocorrelation Φ ′ reflecting the pulse amplitude for each position
(I, j) is as follows.

[0090]

[Equation 12]

[0091]

[Equation 13]

Also for position-weighted autocorrelation reflecting the pulse amplitude for each position, Φ '(j, i) = Φ' (i, j)
There is a relationship. d is a signal or function used for pre-estimation of pulse amplitude for each position. As a typical d,
d = X2 ^t H (cross correlation) can be used. That is,

[Equation 14] And sign (d (i)) is d (i) at the position of n = i
Represents the polarity of. That is, sign (d (i)) and sign (d (j)) are either +1 or -1. The function d used here for estimating the pulse amplitude for each position is an example, and the function d is not limited to this. The position-weighted autocorrelation Φ ′ (i, j) reflecting the pulse amplitude for each position is the pulse amplitude sign (d (i)), sign of position i and position j in the autocorrelation φ (i, j) of the impulse response. Besides multiplying by (d (j)), position i
And the position weights v (i) and v (j) of the position j can be calculated. Position weights v (i) and v (j) are positive values, and v (i) and v (j) are different when position importance is different between position i and position j.
Can have different values (eg, v (i) = 1.25, v (j) = 0.75 depending on the position weight design.
May be).

Further, as a position-weighted pulse amplitude obtained by multiplying the pulse amplitude and the position weight for each position, vs (i) = v (i) × si
Defining gn (d (i)), equation (13) becomes

[Equation 15] Can be written. That is, after the position-weighted pulse amplitude vs (i) is calculated in advance, the position-weighted autocorrelation Φ ′ ((′) that reflects the pulse amplitude for each position by multiplying this by the autocorrelation φ (i, j). The method of obtaining i, j) is also effective. Substituting equation (7) into this,

[Equation 16] Is obtained. If this equation (16) is used, φ (i, j)
The position-weighted autocorrelation can be obtained without using the impulse response and the position-weighted pulse amplitude.

FIG. 6 shows the position weighted correlation calculator 5 of FIG.
11 is a more detailed configuration example of No. 05. In the figure, first, the correlation calculation unit 520 calculates the autocorrelation of the impulse response from the impulse response calculation unit 501, and the position weighting unit 521 reflects the position weight on the calculated autocorrelation, so that the position weighted self The method of obtaining the correlation is shown.

FIG. 7 is an example showing the position weighting unit 521 of FIG. 6 in more detail. The position weighting unit 521 includes an amplitude calculation unit 522, a position weighting unit 523, and a multiplication unit 52.
It is composed of 4. The amplitude calculator 522 calculates the pulse amplitude, and as an example of implementation, d = X2
The pulse amplitude for each position is obtained by the method described above based on ^t H (cross-correlation).

As another method, a method of obtaining the pulse amplitude for each position using the cross-correlation (X3 ^t HV) reflecting the position weight, which will be described later, is also effective. Using this method, the conventional cross-correlation (X2 ^t
The calculation amount is reduced because the calculation of H) becomes unnecessary. The position weighting unit 523 multiplies the position-weighted pulse amplitude by the position weight to obtain the position-weighted pulse amplitude v
Output s (i). By multiplying this by the autocorrelation in the multiplication unit 524, the position-weighted autocorrelation Φ ′ (i, j) that reflects the pulse amplitude for each position is calculated.

Next, position-weighted cross-correlation X3 ^t HV
^{(= R2 t V t H t} HV) An example of Determination will be described.

X3 is X3 = HVr2, and r2 is r2 =
Since it is H ⁻¹ X2, first, the target residual vector r2 is obtained by the equation (6), and then this is multiplied by the position weight to obtain the position-weighted target residual vector Vr2. That is,
If the vector q = Vr2 is set, the element q (n) is

[Equation 17] Can be calculated. Next, the position-weighted target residual vector q
The impulse response is convolved with the position-weighted target vector X3 (= H (Vr2) = Hq).

[0099]

[Equation 18]

Next, the element γ (n) of the vector X3 ^t H is obtained by the correlation calculation between the position-weighted target vector X3 and the impulse response.

[0101]

[Formula 19]

Finally, the element weight γ (n) of X3 ^t H is multiplied by the position weight v (n) to obtain the position weighted cross-correlation X3 ^t HV.
The element f (n) of

[0103]

[Equation 20]

When the pulse amplitude for each position is given, the position-weighted cross autocorrelation f ′ reflecting this is given.
(N) is

[Equation 21] Becomes Here, d (n) is a signal or function used for pre-estimation of the pulse amplitude for each position. As d (n), cross-correlation with no position weight or cross-correlation with position weight f
(N) can be used. That is,

[Equation 22]

FIG. 8 is a more detailed configuration example of the position-weighted cross-correlation calculation unit 506 of FIG. In the figure, the position-weighted cross-correlation calculation unit 506 includes a position-weighted target vector calculation unit 525, a cross-correlation calculation unit 529, and a position weighting unit 530. The position-weighted target vector calculation unit 525 calculates the position-weighted target vector based on the impulse response, the position weight, and the target vector from which the influence of the adaptive codebook search unit 504 is subtracted.

Next, the cross-correlation calculator 529 calculates the cross-correlation between the impulse response and the position-weighted target vector,
A position weighting unit 530 multiplies this by a position weight to obtain a position weighted cross-correlation.

FIG. 9 is an example showing the position-weighted target vector calculation unit 525 of FIG. 8 in more detail. The position-weighted target vector calculation unit 525 includes a target residual vector calculation unit 526, a position weighting unit 527, and a filtering unit 528. Target residual vector calculation unit 5
Reference numeral 26 calculates a target residual vector using the target vector and impulse response from which the influence of the adaptive codebook search unit 504 has been subtracted. The position weighting unit 527 multiplies the target residual vector by the position weight and outputs the position-weighted target residual vector. Next, the filtering unit 528 convolves the impulse response with the position-weighted target residual vector to obtain the position-weighted target vector.

The gist of the present invention is to perform a codebook search using the distortion evaluation value calculated using the position weight and the impulse response, and based on the position weighted correlation and the position weighted cross correlation. The code search method is a method for realizing the method of the present invention with a small amount of calculation. Therefore, although there are other methods similar to the method of the present embodiment and a method of simplifying the calculation, it goes without saying that such a case is basically included in the present invention.

For example, a position-weighted cross-correlation is obtained.
As another method, there is the following method. Ie position
Weighted cross-correlation X3^tHV is r2^tV^tH^tHV etc.
The target residual vector r2 and the above-mentioned position
Weighted autocorrelation Φ (i, j) (= V^tH^tThe point of HV
Position weighted cross-correlation X3
^tThe element f (n) of HV is

[Equation 23] Can be calculated using

Compared with the calculation of the equation (9) of the conventional method in the order of the cube of L, the increase in the amount of calculation according to the present invention is at most L regardless of which of the above-mentioned methods is used.
It can be said that the method of the present invention is a practical method.

The above is the description of the method of calculating the position-weighted autocorrelation and the method of calculating the position-weighted cross-correlation of the present invention.

Now, returning to FIG. 3, the description will be continued.

The random codebook search unit 507 searches the code k of the code vector with the minimum distortion using the algebraic codebook. At this time, the position weighted correlation and the position weighted cross correlation are used. The algebraic codebook is a codebook in which the values that can be taken by the amplitude of a predetermined Np pulse are limited to +1 and -1, and a code vector is represented by a combination of pulse position information and amplitude information (that is, polarity information). . The characteristics of the algebraic codebook are that it is not necessary to directly store the code vector itself, so the amount of memory representing the codebook is small, and although the amount of calculation for selecting the code vector is small, it is relatively small. It is possible to represent the noise component included in the sound source information with high quality. ACEL is one that uses an algebraic codebook for encoding excitation signals in this way.
It is known that the P system and the ACELP-based system are available, and that synthetic speech with relatively little distortion can be obtained.

Under such a structure, the random codebook search unit 5
In 07, a synthesized speech signal reproduced using an output signal (code vector) from the codebook and a target signal that is a target in the noise codebook search unit (modified target vector X2)
Distortion is evaluated using a position weighted correlation (V ^t H ^t HV) and a position weighted cross correlation (X3 ^t HV), and an index (noise code) C that reduces the distortion is searched for. The evaluation value used at this time is, as described above,
The distortion evaluation value of the second term on the right side of the equation (8) is (X3 ^t HVc k) ² / (c k ^t V ^t H ^t HVc k), and searching for the code k of the code vector that maximizes this value is the most distortion. This is equivalent to selecting a code that reduces. Then, the searched random code C and the random code vector c _c corresponding to this code are output.

When the random codebook is realized by an algebraic codebook, it is composed of several (Np in this case) non-zero pulses, so that the second term on the right side of the equation (8) X3 ^t HVcc on the numerator side of the evaluation value is further

[Equation 24] It can be expressed as. Where m _i is the position of the i-th pulse,

[Equation 25] Is the amplitude of the i-th pulse, and f (n) is an element of the position-weighted cross-correlation vector X3 ^t HV. Also,
Ck ^t V ^t H on the denominator side of the evaluation value of the second term on the right side of the equation (8)
^t HVcc is

[Equation 26] It can be expressed as. Furthermore, under the condition that the pulse amplitude for each position is obtained, using the position-weighted correlation that reflects this

[Equation 27]

[Equation 28] Becomes Based on these, distortion evaluation value (X3 ^t HVcc) ² /
^{(Ck t V t H t HVck} ) Selection of pulse position information that explore the maximum to become such a pulse position m _{i (i} = 0~Np) is completed. At this time, f '(n) and Φ' (i,
Since j) (or f (n) and Φ (i, j)) is calculated before the search, the calculation amount required for code selection is very small. The selected pulse position information is output as a noise code together with the pulse amplitude information.

The present invention has the effect of more effectively selecting a code in which distortion is less audible by combining position weighting and conventional auditory weighting. In order to use the auditory weighting, as described above, the impulse response h
It is necessary to obtain (n) from the perceptual weighted synthesis filter H (z) as shown in the equation (1).

The gain codebook search unit 509 is used to express the gain component of the excitation. In a typical CELP method, the gain codebook search unit 509 encodes two types of gains, a gain used for a pitch component and a gain used for a noise component. Here, for the sake of simplicity of explanation, the gain codebook searching unit 509 will be described as a configuration that does not use the position weight, but it goes without saying that the position code weight may be used in the gain codebook searching unit 509.

In the gain codebook search, an index (gain code) G is searched so that the distortion between the synthesized voice signal reproduced using the gain candidates extracted from the codebook and the target voice signal becomes small. Then, the searched gain code G and the corresponding gain are output.

Although the method of using the position weights only for the random codebook search has been described here, it is needless to say that the present invention is not limited to this and various modifications can be made. For example, a method of using position weights for each of the three codebook searches in the adaptive codebook search, the noise codebook search, and the gain codebook search is also effective.

In another implementation, it is also effective to use position weights only for the two search units of the adaptive codebook search and the random codebook. In yet another implementation, a method of using the position weight only for the gain codebook search is also effective.

As described above, the present invention can be applied in various forms. In any usage, the position weight obtained from the voice signal is used as the time series signal of the residual signal (or sound source signal) level. By using it for encoding, signal samples at important positions can be encoded more accurately.

Excitation signal generation section 510 has adaptive code vector and adaptive codebook search section 5 from adaptive codebook search section 504.
Noisy code vector from 07, gain codebook search unit 50
The sound source signal is generated using the gain from 9. The generated excitation signal is stored in the adaptive codebook so that it can be used by adaptive codebook searching section 504 in the next coding section. Further, the generated excitation signal is stored in the adaptive codebook so that it can be used by adaptive codebook searching section 504 in the next coding section. The generated sound source signal is the target signal calculation unit 5
At 11, it is used to calculate the target signal for encoding in the next interval.

This is the end of the description of the speech coding shown in FIG.

FIG. 10 is a flowchart of the encoding method according to the embodiment of the present invention. A voice signal is input for each predetermined coding section (step S1), spectrum parameters are analyzed and a codebook search is performed (step S2). Next, a target signal for encoding the current section is calculated (step S3). Next, the impulse response is obtained based on the spectrum parameter (step S4).

Further, a first signal (eg residual signal) is obtained from the voice signal (step S5), and the position weight is set using the first signal (step S6). As an example, the position weight is determined through shape information that does not depend on the polarity, such as the absolute value amplitude of the residual signal and the outer shape of the power of the residual signal. Then, an adaptive codebook search is performed (step S7).

Next, the position-weighted correlation and the position-weighted cross-correlation are calculated (step S8), the random codebook is searched using these, and the random code is selected (step S).
9). Next, the gain codebook is searched (step S1).
0). The spectrum parameter code, adaptive code L, noise code C, and gain code G thus obtained are output from the encoding unit (step S11), and the excitation signal is calculated in preparation for the encoding of the next section (step S12). If the next section is to be coded, the audio signal of the next section is input in step S1, and if not, the processing of the coding unit ends in step S13.

This is the end of the description of the processing of the encoding unit using the flowchart of FIG.

Since the present invention relates to weighting used for code selection of parameters performed on the encoding side, the decoding method may be the same as the conventional method. Here, the decoding method will be briefly described with reference to FIG.

In FIG. 11, the coded data from the coding unit is input from the input terminal 160, and is separated into each code A, L, C, G in the coded data separation unit 19. The spectrum parameter decoding unit 14 reproduces the spectrum parameter based on the code A. The adaptive excitation decoding unit 11 reproduces an adaptive code vector based on the code L. The noise source decoding unit 12 reproduces a noise code vector based on the code C.
The gain decoding unit 13 reproduces the gain based on the code G. In the sound source reproducing unit 15, the reproduced adaptive code vector,
The sound source signal is reproduced using the noise code vector and the gain.

The synthesizing filter 16 forms a synthesizing filter by using the spectrum parameters reproduced by the spectrum parameter decoding unit 14, and passes the sound source signal from the sound source reproducing unit 15 through this to generate a synthetic speech signal. . The post filter 17 performs post filtering processing for shaping the coding distortion included in the synthesized voice signal so that the sound becomes easy to hear. The processed synthesized voice signal is output from the output terminal 195.

This is the end of the description of the present embodiment.

[0132]

According to the present invention, it is possible to provide an audio / audio signal encoding method and an electronic device capable of generating a high-quality audio / audio signal even at a low bit rate.

[Brief description of drawings]

FIG. 1 is a diagram showing a principle configuration of residual level position weighting in an encoding method of the present invention.

FIG. 2 is a diagram showing a principle configuration of a coding method of the present invention for performing code selection of a random codebook by using position weighting of residual signal levels and distortion evaluation at a perceptually weighted synthesis level.

[Fig. 3] Fig. 3 is a block diagram showing an example of a coding unit that realizes a method of coding a voice / acoustic signal of the present invention.

FIG. 4 is a diagram (No. 1) for explaining an example of a method for obtaining a position weight in a position weight setting unit 503.

FIG. 5 is a diagram (No. 2) for explaining an example of a method for obtaining a position weight in a position weight setting unit 503.

6 is a diagram showing a more detailed configuration example of a position weighted correlation calculation unit 505 in FIG.

7 is a diagram showing an example of the position weighting unit 521 of FIG. 6 in more detail.

8 is a diagram showing a more detailed configuration example of a position-weighted cross-correlation calculation unit 506 in FIG.

9 is a position weighted target vector calculation unit 525 of FIG. 8;
It is a figure which shows an example which represented in more detail.

FIG. 10 is a flowchart showing an encoding method according to an embodiment of the present invention.

FIG. 11 is a diagram for explaining a decoding method.

FIG. 12 is a diagram showing a process in which an auditory weighted combined speech signal is generated from a sound source signal by auditory weighted synthesis.

[Explanation of symbols]

300 sound source signal (residual level) 302 Residual Level Position Weighting Unit 303 Auditory weighted synthesis unit 308, 309 route 310 output terminal 500 Spectral parameter analysis / encoding unit 501 impulse response calculator 502 Residual signal calculation unit 503 Position weight setting unit 504 Adaptive codebook search unit 505 Position weighted correlation calculator 506 Position-weighted cross-correlation calculator 507 Random codebook search unit 509 Gain Codebook Search Unit 510 sound source signal generator 511 Target signal calculation unit 900 spectrum parameter processing unit 901 Target residual signal generation unit 902 Noise Codebook 903 Residual level position weight acquisition unit 904, 905 Residual level position weighting unit 906, 907 Auditory weighted synthesis unit 908 Gain multiplication unit 909 Strain evaluation unit 910 code selection unit

Claims (10)

[Claims] 1. A method of encoding a voice / acoustic signal including encoding of pulse position information, the method comprising: acquiring a parameter representing a short-time spectrum of an input signal; And a step of obtaining at least position weight information from an input signal, and a step of selecting the pulse position information using the impulse response and the position weight information. A method for encoding a voice / sound signal. 2. A method for encoding a voice / acoustic signal including encoding of pulse position information, the method comprising: acquiring a parameter representing a short-time spectrum of an input signal; A step of obtaining an impulse response, a step of obtaining at least position weight information from an input signal, a step of calculating a position weighted correlation value using the impulse response and the position weight information, and a calculated position weighted correlation value And a step of selecting the pulse position information by using the above method. 3. A method for encoding a voice / acoustic signal including encoding of pulse position information, the method comprising: acquiring a parameter representing a short-time spectrum of an input signal; A step of obtaining an impulse response, a step of obtaining at least a target signal and position weight information from an input signal, a step of obtaining a position weighted correlation value based on the impulse response and the position weight information, the impulse response and the A step of obtaining a position weighted cross-correlation value using the target signal and the position weight information; and a step of selecting the pulse position information using the position weighted correlation value and the position weighted cross correlation value. A method for encoding a voice / acoustic signal, comprising: 4. The method of encoding a voice / acoustic signal according to claim 1, wherein an algebraic codebook is used for encoding the pulse position information. 5. The method according to claim 1, further comprising the step of obtaining a short-term prediction residual signal from the input signal, wherein the position weight information is obtained using the short-term prediction residual signal.
5. The audio / audio signal encoding method according to any one of 1 to 4. 6. A method of encoding a voice / acoustic signal including encoding of information representing a sound source signal, comprising: acquiring a parameter representing a short-time spectrum of an input signal; and a parameter representing the acquired short-time spectrum. A step of obtaining an impulse response based on, a step of obtaining position weighting information using a short-term prediction residual signal obtained from an input signal, and using the impulse response and the position weighting information, information representing the sound source signal The method for encoding a voice / acoustic signal, comprising: 7. The method for encoding a voice / acoustic signal according to claim 1, wherein both position weighting and auditory weighting are used. 8. A method for encoding a voice / acoustic signal including encoding of information representing a sound source signal, the method comprising: position-weighting a candidate signal for representing a sound source signal, the position-weighted candidate signal A method of encoding a voice / acoustic signal, comprising the step of selecting the candidate signal based on the distortion when the above are combined. 9. A method for encoding a voice / acoustic signal including encoding of information representing a sound source signal, the method comprising: position-weighting a candidate signal for representing a sound source signal; and the position-weighted candidate signal. A method of encoding a voice / acoustic signal, comprising the step of selecting the candidate signal based on the distortion when the sound weighted synthesis is performed. 10. An input unit for inputting a voice / acoustic signal, an encoding unit for performing an encoding process on the voice / acoustic signal input via the input unit, and an encoding unit for encoding by the encoding unit. A transmitting unit for transmitting the encoded audio / acoustic signal, a receiving unit for receiving the encoded audio / acoustic signal, and a decoding process for the audio / acoustic signal received via the receiving unit. A decoding unit and an output unit for outputting the audio / acoustic signal decoded by the decoding unit are provided, and the coding unit is the coding method according to any one of claims 1 to 9. An electronic device characterized by being executed.