JPH10105200A - Voice coding/decoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a synthesis voice of high quality even with a low bit rate by coding a driving signal using a Formant emphasizing filter having a characteristic based on a synthesis filter, plural driving signal candidates, and an input voice signal. SOLUTION: A synthesis filter information coding section 100 extracts information of a spectrum envelope of an input voice signal, and gives parameters of the synthesis filter to a Formant emphasis information generating section 102. A weight filter 101 performs weighting of an input voice signal. A pitch component coding section 105 encodes a pitch driving signal of the synthesis filter. A Formant emphasizing section 107 generates a candidate of a formant- emphasized noise driving signal using Formant emphasizing information from the Formant emphasis information generating section 102. A noise driving signal searching section 108 searches a candidate in which distortion is reduced, and outputs a noise driving signal and a weighting synthesis noise signal respectively Foltman-emphasized are outputted as an output of a noise component coding section 109.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to telephone band voice,
The present invention relates to a speech encoding / decoding method for performing compression encoding / decoding of a wideband speech or an audio signal.

[0002]

2. Description of the Related Art As a technique for compression-encoding an audio signal,
CELP (Code-Excited Linear Prediction) is known. For details of the CELP method, see MRS
chroeder, and BSAtal “Code-Excited Linear Predic
tion (CELP): High-quality Speech at Very Low Rates ”
Proc. ICASSP '85, 25, 1.1, pp. 937-940, 1985 (Reference 1)
Is disclosed.

[0003] The CELP system is a typical one of voice coding systems called analysis-synthesis coding. In the analysis-synthesis type coding, a speech signal is modeled by a synthesis filter and a drive signal for driving the synthesis filter, and the drive signal is passed through the synthesis filter to generate a synthesized speech signal. Then, the characteristics of the synthesis filter and the drive signal are encoded and output. A filter using an all-pole filter as a synthesis filter is called an LPC (Linear Predictive Coding) synthesis model in the field of speech coding and speech synthesis. The synthesis filter represents the resonance characteristics of the vocal tract, and the drive signal corresponds to the vocal cord signal. The resonance point of the vocal tract is called a formant.

In the conventional analysis-synthesis type coding, a noise drive signal is selected from previously prepared candidates for a static white noise signal, and a pitch drive signal having a pitch period corresponding to the pitch of a voice is combined with the selected noise drive signal. And the final drive signal. C
In the ELP method, in order to generate or reproduce this static noise signal, a noise codebook that stores noise signal candidates determined in advance in design is often used.

[0005] A human voice is used for a short time (5 to 10 msec).
Since the information of the sound source and the vocal tract is relatively small in the section c), by updating the drive signal and the information of the vocal tract for each section, the LPC model is relatively used under a relatively high bit rate. Voices can be expressed efficiently. The CELP method further evaluates the encoding of the drive signal based on the distortion of the audio signal level weighted by the auditory sense. Therefore, if the bit rate is up to about 8 kbit / s,
A decoded voice with little distortion can be obtained.

However, if the number of bits used for encoding is reduced and the bit rate is reduced to about 4 kbit / s, the distortion due to encoding becomes apparent as a sound especially due to the lack of bits for expressing the drive signal. It becomes perceived, and deterioration such as lack of naturalness or noise mixing becomes remarkable.

[0007]

As described above, in a conventional speech coding method based on a speech synthesis model such as the CELP system, when the bit rate is reduced to about 4 kbit / s, coding distortion is clearly perceived. And it becomes difficult to obtain high quality decoded speech. An object of the present invention is to provide a speech encoding / decoding method capable of obtaining a high-quality synthesized speech even at a bit rate of about 4 kbit / s.

[0008]

In order to solve the above-mentioned problems, the present invention expresses a synthesized speech signal using a synthesis filter and a drive signal for driving the synthesis filter, and distorts the synthesized speech signal with respect to the input speech signal. In a speech encoding method for encoding information and a drive signal representing characteristics of a synthesis filter that reduces the size of a synthesis filter, a synthesis filter, a plurality of drive signal candidates, and a formant emphasis filter having characteristics obtained based on an input audio signal are provided. It is a basic feature that coding of the drive signal is performed by using this.

More specifically, the distortion of the synthesized voice signal is reduced by using a plurality of drive signal candidates including the drive signal candidates which are formant-emphasized by the formant emphasis characteristics obtained based on the input voice signal. A drive signal is selected, and the selected drive signal is encoded.

When the driving signal is expressed by a combination of a pitch driving signal and a noise driving signal, at least a part of at least one of the pitch driving signal candidate and the noise driving signal candidate is formant-emphasized.

In this way, a suitable drive signal can be selected from the drive signal candidates that have been adapted by dynamic formant enhancement in accordance with the spectral characteristics of the voice, so that the conventional static drive signal candidates can be selected. As compared with the selection method, a synthesized speech signal with significantly smaller coding distortion can be generated.

Further, the present invention is characterized in that a formant emphasis characteristic is obtained from information indicating characteristics of a synthesis filter, and formant emphasis of a drive signal candidate is performed. In this way, by transmitting information representing the characteristics of the synthesis filter from the encoding side to the decoding side, it is possible to transmit special side information for formant enhancement from the encoding side to the decoding side without transmitting the side information for encoding. A common formant enhancement can be performed on both the decoding side and the decoding side. That is, it is possible to greatly improve the quality of synthesized speech without increasing the amount of information.

Further, in the present invention, the formant emphasizing characteristic may include a characteristic for correcting a spectrum inclination.
By providing such a spectrum inclination correction characteristic to the formant emphasis characteristic, it is possible to prevent an unnecessary spectrum inclination mixed in the formant emphasis processing from being given to the drive signal. As a result, it is possible to prevent a phenomenon in which the synthesized speech is muffled in time, and to obtain a more stable synthesized speech.

The following is another reason why the formant enhancement of the drive signal candidate is effective.
Speech analysis and synthesis filter information (representing phoneme information)
When the drive signal (drive signal before encoding) obtained by removing the audio signal from the input audio signal is reproduced as audio, the content of the audio can often be heard to some extent. This means that some of the phoneme information that should have been removed remains in the drive signal, and the drive signal is not completely whitened.
For this reason, by applying a formant enhancement process to the drive signal candidates when the drive signal is encoded, a drive signal that is closer to the actual one can be expressed.

Also, since formant enhancement has the function of reducing the frequency components in the valleys between speech formants, the technique of the present invention in which formant enhancement is used as a search candidate for a drive signal on the encoding side is used. It is possible to select a more suitable drive signal that reduces the distortion of the synthesized speech signal from among those with reduced frequency components unrelated to formants, so that the subjective quality of the final synthesized speech itself is improved. effective. For the above reasons, according to the present invention, the number of bits is small, for example, 4 kbi.
Even at a low bit rate of about t / s or less, high-quality synthesized speech can be generated on the decoding side.

Further, in the present invention, when encoding a noise drive signal, the characteristics of a perceptually weighted synthesis filter, the characteristics of a formant emphasis filter, and the target vector used for encoding the drive signal are determined based on the input speech signal. A formant-enhanced impulse response is obtained by using these auditory weighted synthesis filter and formant-enhancement filter, a corrected target vector is obtained by using the target vector and the formant-enhanced impulse response, and a noise drive signal candidate and a corrected target vector are obtained. The noise drive signal may be encoded using the impulse response with formant emphasis. In this case, the characteristics of the pitch emphasis filter may be further obtained, and the impulse response with formant emphasis may be obtained using a synthesis filter with auditory weights, a pitch emphasis filter, and a formant emphasis filter.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. (First Embodiment) FIG. 1 is a block diagram showing a configuration of a speech encoding / decoding device in which an improved synthesis model according to a first embodiment of the present invention is incorporated in a conventional CELP system. It is. The present embodiment differs from the prior art in the processing required for formant emphasizing the drive signal, and the other parts have the same configuration as the conventional CELP system. In the present embodiment, an example of a coding configuration in which formant emphasis is applied particularly to noise drive signal candidates will be described.

[Encoding side] First, the encoding side will be described. On the encoding side, a synthesis filter information encoding unit 100, a weight filter 101, a formant emphasis information generation unit 102, a pitch component encoding unit 105, a noise component encoding unit 109, a gain component encoding unit 110, and a local decoding unit 111 are provided.

The synthesis filter information coding unit 100 analyzes a speech signal (hereinafter, referred to as an input speech signal) input from the input terminal 11 in units of a fixed length called a frame, and indicates a short-term correlation of the speech signal. The information of the spectral envelope is extracted and encoded. The information encoded by the synthesis filter information encoding unit 100 is provided to the multiplexing unit 112 as a code representing the characteristics of the synthesis filter, and is further converted to parameters of the synthesis filter and provided to the formant emphasis information generation unit 102. As a method of analyzing a speech signal, for example, linear prediction analysis (LPC: Linear Predi
ction Coding) can be used.

The weighting filter 101 is a filter used to improve subjective sound quality by utilizing the masking characteristics of the auditory sense.
A transfer function is set based on the information of the spectral envelope analyzed by the above, and the weighting of the input speech signal is performed, and a target which is a target of the coding is removed except for the influence of the past coding from the local decoding unit 111. Generate a vector.

The pitch component encoding section 105 encodes a pitch cycle component (referred to as a pitch drive signal) of the drive signal for driving the synthesis filter, and includes an adaptive codebook 103 and a pitch drive signal search section. 104.

The adaptive codebook 103 stores past encoded drive signals and generates pitch drive signal candidates representing pitch period components included in the target vector input from the weight filter 101. The pitch driving signal search unit 104 generates a weighted synthesized pitch signal obtained by synthesizing a pitch driving signal by a weighted synthesis filter with respect to a pitch cycle candidate determined by a predetermined method and a corresponding pitch driving signal candidate. A candidate that reduces distortion (error of a weighted synthesized pitch signal with respect to a target vector) is searched for. The pitch cycle selected in this manner, the pitch drive signal e ₁ (n) corresponding to the pitch cycle, and the weighted synthesized pitch signal are output from the pitch component encoding section 105.

Next, the noise component coding unit 109 to which formant enhancement is applied according to the present invention will be described in detail. The noise component coding unit 109 includes a noise drive signal candidate generation unit 106, a formant enhancement unit 107, and a noise drive signal search unit 104.

The noise drive signal candidate generating section 106 is configured to generate a noise drive signal candidate that can be uniquely decoded in association with a noise code by a predetermined method.
The noise drive signal candidate generating unit 106 can be generally realized by a statically provided noise codebook, but is not limited to this. A combination of a plurality of small noise codebooks or a combination in advance is limited. Alternatively, a signal equivalent to a static noise codebook may be expressed by a pulse or the like. The present invention can cope with all the driving signal candidates generated in this way, and by performing formant emphasis on them, a part or all of the noise driving signal candidates can be deformed so that coding distortion can be dynamically suppressed. .

The formant emphasis unit 107 generates a formant-enhanced noise drive signal using the formant emphasis information from the formant emphasis information generation unit 102. Using the parameters of the synthesis filter from the synthesis filter information encoding unit 100, the formant enhancement information generation unit 102 obtains formant enhancement information that reflects the relationship between the unevenness of the spectral envelope of the synthesis filter.

FIG. 2 schematically shows the relationship between the synthesis filter according to the present invention and the formant enhancement characteristics. FIG. 2A shows an example of the spectral characteristics of the synthesis filter, and FIG.
Shows an example of the formant enhancement characteristic according to the present invention corresponding to the characteristic of the synthesis filter. In FIG. 2A,
The synthesis filter takes the peak value of the power spectrum at the frequencies (angular frequencies) ω1, ω2, ω3 (ω1, ω
The frequency between 2 and ω3 is a valley), and it can be considered that these peaks correspond to the formants.
In addition, in this example, it can be seen that the spectrum has a low-pass characteristic which is inclined downward to the right as a whole.

On the other hand, the formant emphasis characteristic in FIG. 2B has a peak value at ω1, ω2, ω3 corresponding to the frequency position indicating the peak of the spectrum of the synthesis filter. However, the formant emphasis characteristic is a flat characteristic with no spectrum inclination as a whole so as not to bring an unnecessary spectrum inclination to the drive signal.

FIG. 2C shows an example of drive signal candidates when formant enhancement is performed according to the characteristics shown in FIG. 2B. On the other hand, FIG. 2D shows an example of a conventional drive signal candidate without formant enhancement. As can be seen by comparing FIGS. 2 (c) and 2 (d), by performing a code search of the drive signal using the drive signal candidate when the formant is enhanced according to the present invention, a clear synthesized speech with a clear formant is generated. It is possible to select a drive signal that performs the following.

An example of a specific method for realizing such a formant enhancement characteristic will be described below. here,
Formant emphasis characteristic Hc (z) is converted to synthesis filter H
It is determined from the information of (z). That is, it can be realized by the following equation: Hc (z) = Hf (z) H (z / ν1) / H (z / ν2) (1) Here, Hf (z) is Hc
A filter for flattening the slope of the spectrum of (z) is shown. Further, ν1 and ν2 are parameters for adjusting the degree of unevenness of Hc (z), and it is desirable that the characteristic is close to the unevenness of the synthesis filter and that the peak appears gently. It is necessary that ν1> ν2 ≧ 0. Hf (z) is H (z / ν1) / H (z / ν
It can be obtained by flattening the overall slope of the spectrum of 2). As a specific example,
The values R1,..., Rq corresponding to the autocorrelation coefficients of the impulse response of H (z / ν1) / H (z / ν2) are obtained, and based on the values, the prediction coefficient of the q-th order is calculated in the same manner as in the LPC analysis. Asked,
A prediction filter using this prediction coefficient is defined as Hf (z). The order q of prediction uses a value of 1 or more.

When the value of the prediction order q is increased to about 2 to about 4 or more, low-order formants other than the slope of the entire spectrum are flattened, and special formant emphasis such as emphasizing only high-order formants is performed. It is possible to do. When the higher-order formants are emphasized, there is an effect that the phoneme becomes clear and the clarity of the synthesized speech can be increased.

FIG. 3 shows an example of the relationship between the synthesis filter according to the present invention and the characteristic of formant enhancement when q having a value of 4 or more is used. FIG. 3A shows an example of the spectrum characteristic of the synthesis filter, and FIG. 3B shows an example of the formant enhancement characteristic according to the present invention corresponding to the synthesis filter. In FIG. 3B, unlike the case where q having a small value described above is used, the formant emphasis characteristic is only in the vicinity of ω3 corresponding to the third formant among the frequency positions ω1, ω2, ω3 indicating the peaks of the spectrum of the synthesis filter. It takes a peak value and the other parts are flattened. As shown in FIG. 3C, an example of a spectrum of a drive signal candidate when formant emphasis is performed by using this is a characteristic in which a peak value is obtained only near ω3 corresponding to the third formant in many cases. . Therefore, there is an effect that a high-order formant, which is difficult to sufficiently express in the conventional CELP method, is emphasized, and a synthesized speech with clear phonemes can be generated. When a synthesis filter obtained based on LPC analysis is used, the synthesis filter can be expressed by the following equation (2).

[0032]

(Equation 1)

P is the order of the synthesis filter. Therefore,
In this case, the formant emphasis characteristic is given by the following equations (1) and (2).
Thus, Hc (z) = Hf (z) A (z / ν2) / A (z / ν1) (3)

The formant emphasis characteristic Hc (z) can be reproduced on the decoding side using the information of the synthesis filter that can be reproduced on the decoding side. Therefore, with a configuration in which the code of the synthesis filter and the noise code are transmitted from the encoding side, it is possible to reproduce a drive signal in which the same formant as that of the encoding side is emphasized on the decoding side.

Further, in order to reduce the calculation amount of the formant enhancement processing, a method is used in which the window of the impulse response of Hc (z) is windowed to cut off M + 1 samples, and this is used as a coefficient of an M-order FIR filter for formant enhancement. It is valid. A rectangular window, a Hamming window, an exponential window, or the like can be used as the window for censoring.

Now, the noise driving signal before formant emphasis is e ₂ (n), and an M-order FIR used for formant emphasis.
Assuming that the impulse response of the filter is h _c (n), the noise drive signal e ₂ ′ (n) with formant emphasis can be obtained by, for example, a convolution process represented by the following equation.

[0037]

(Equation 2)

Here, h _c (0) = 1. Referring back to FIG. 1, the noise driving signal search unit 108 generates a noise driving signal candidate obtained by synthesizing a formant-emphasized noise driving signal candidate with a weighted synthesis filter such that the distortion becomes smaller with respect to the target vector excluding the pitch component. And outputs the selected noise code, the corresponding formant-enhanced noise drive signal and the formant-enhanced weighted combined noise signal as the output of the noise encoding unit 109.

Although a noise code can be selected by the above method, there is another method of searching for a noise code which has the same principle but a different method. This is because the combination of the weighted synthesis filter and the formant emphasis filter is regarded as a search weighted synthesis filter again without performing the formant emphasis processing of the noise drive signal candidate, and the noise drive signal candidate before formant emphasis is regarded as a search. This is a method of searching for a noise code using the above method. This is also equivalent to formant emphasis of the noise drive signal, so that the same noise code can be selected in the search result. By using this method, it is not necessary to perform formant emphasis on the noise drive signal candidates each time, and the formant emphasis processing can be eliminated from the noise drive signal search loop. Therefore, the amount of calculation required for the search can be significantly reduced.

Next, the gain component encoder 110 will be described. The gain component encoding section 110 encodes a gain value used for each of the pitch drive signal and the noise drive signal. The gain value is encoded by a gain that gives a gain such that a synthesized signal obtained by combining both the weighted synthesized pitch signal when multiplied by the gain and the weighted synthesized noise signal subjected to formant emphasis is close to the target signal. This is done by selecting the code and outputting it.
Further, a drive signal to be input to the synthesis filter is obtained using the gain, the pitch drive signal e ₁ (n), and the formant-emphasized noise drive signal e ₂ ′ (n).

Ex (n) = g ₁ e ₁ (n) + g ₂ e ₂ ′ (n) (5) where g ₁ and g ₂ represent a gain for the pitch drive signal and a gain for the noise drive signal, respectively. .

The drive signal thus obtained is stored in the adaptive code book 103 in preparation for encoding the next pitch component. Finally, the drive signals are weighted and combined to determine the effect on the encoding of the next frame, and the combined signal generated as a result is input to the weight filter unit 101.

The coded data (coded data) is multiplexed by the multiplexing unit 112 and transmitted as a bit stream to the decoding side. [Decoding Side] Next, the decoding side will be described. In FIG. 1, the bit stream sent from the encoding side is first input to the demultiplexing unit 113 on the decoding side. The demultiplexing unit 113 separates the input bit stream into a synthesis filter code, a pitch cycle code, a noise code, and a gain code, and organizes these into codes for each frame and outputs them.

On the decoding side, a synthesis filter information reproduction unit 114, a formant emphasis information generation unit 115, an adaptive codebook 116, a noise drive signal reproduction unit 117, a formant emphasis unit 119, a gain component reproduction unit 118, a drive signal generation unit A unit 120 and a synthesis filter 121 are provided, and reproduce a synthesized audio signal in frame units based on the code separated by the demultiplexing unit 113. Hereinafter, each part will be described in detail.

The synthesis filter information reproducing section 114 reproduces the information of the synthesis filter using the code of the synthesis filter. The adaptive codebook 116 stores a past pitch drive signal, and a pitch drive signal is extracted from the adaptive codebook 116 based on a pitch cycle code.

The formant emphasis information generation section 115 has the same function as the formant emphasis information generation section 102 described on the encoding side, and reflects the relationship between the unevenness of the spectrum envelope of the synthesis filter based on the decoded synthesis filter information. Parameter information for formant emphasis as described below is obtained.

The noise drive signal reproducing section 117 reproduces the noise drive signal based on the noise code in the same manner as the encoding side.
The formant emphasis unit 119 has the same function as the formant emphasis unit 107 on the encoding side, and performs formant emphasis of the noise drive signal using the parameter from the formant emphasis information generation unit 115. In order to generate the same formant-enhanced drive signal on the encoding side and the decoding side, the formant emphasis unit 119 needs to perform the same or equivalent formant emphasis as the formant emphasis unit 107 on the encoding side. In addition, information different from the information of the synthesis filter and obtained from the decoding side, such as voiced
Depending on the information on the degree of silence, etc., the degree of formant emphasis can be further increased or reduced on the encoding side.

The gain component reproducing section 118 obtains a gain to be given to each of the pitch drive signal and the formant-emphasized noise drive signal based on the gain code. The drive signal generator 120 generates a drive signal to be input to the synthesis filter by combining the pitch drive signal and the noise drive signal subjected to formant emphasis by giving the gain reproduced by the gain component reproducer 118. The generated drive signal is stored in the adaptive codebook 116.

The synthesis filter 121 is a filter whose characteristics are determined by the synthesis filter information reproduced by the synthesis filter information reproduction section 114. The synthesis filter 121 generates a synthesized voice signal by inputting a drive signal from the drive signal generation section 120. To the output terminal 12. In order to further improve the sound quality, the synthesized voice signal may be output after passing through a post filter.

Next, the processing procedure on the encoding side in this embodiment will be described with reference to the flowchart shown in FIG. Normally, in the CELP method, synthesis filter information is performed in units of frames, and encoding of other pitch periods, noise drive signals, and gains is often performed in units of subframes obtained by further dividing a frame. ,
An example of a case where the coding of the pitch period, the noise drive signal, and the gain is designed to be performed in frame units will be described.

First, initial settings are made (step S10).
0), and then input a sound for frame processing (step S).
101), the information of the synthesis filter is extracted and encoded (step S102). Next, the input audio signal is weighted by a weight filter obtained from the analyzed information on the spectrum envelope, and a target signal to be a target of encoding of the drive signal is generated (step S103). Further, a search for a pitch drive signal is performed, and the selected pitch cycle is encoded (step S104). Then, using the information of the synthesis filter for formant enhancement, formant enhancement information that reflects the relationship between the unevenness of the spectrum envelope of the synthesis filter is obtained (step S105). Next, formant-emphasized noise drive signal candidates are generated, and the noise drive signal is searched for and encoded using these candidates (step S).
106). Next, gain values used for the pitch drive signal and the noise drive signal are encoded (step S10).
7).

The coded data (coded data) is multiplexed and transmitted as a bit stream to the decoding side (step S108). Further, a drive signal is generated by giving a gain, and local decoding is performed to generate a synthesized speech signal with auditory weight (step S109). Finally, processing for the next frame is provided, and internal data used for encoding is moved (step S110).

Next, the processing procedure on the decoding side in this embodiment will be described with reference to the flowchart shown in FIG. First, the internal state of the decoding process is initialized (step S
200). Next, the bit stream (coded data) sent from the coding side is separated into codes used for decoding (step S201). Among these, the code of the synthesis filter is used to reproduce the synthesis filter information (step S202), and the pitch period code is used to reproduce the pitch drive signal (step S20).
3). Further, the information of the synthesis filter is also used to obtain formant emphasis information (step S204),
Formant enhancement of the noise drive signal is performed using the noise drive signal reproduced with the noise code and the formant enhancement information (step S205). Next, gain information is reproduced using the gain code (step S206), and the driving signal is obtained by combining the pitch driving signal and the formant-emphasized noise driving signal using the reproduced gain (step S207). Then, the drive signal is input to a synthesis filter to create a synthesized voice signal (step S208).
Finally, the internal data is moved in preparation for the processing of the next section (step S209).

(Second Embodiment) FIG. 6 is a block diagram showing a configuration of a speech encoding / decoding device according to a second embodiment.

In the first embodiment, the configuration in which the formant enhancement is performed on all the noise drive signal candidates has been described. However, in the second embodiment, the formant enhancement is performed particularly on a part of the noise drive signal candidates. The configuration when applied to candidates will be described. Therefore, the parts other than the part related to the processing of the noise drive signal can be basically realized in the same manner as in the first embodiment. In FIG. 6, the same processing is denoted by the same reference numeral as in FIG. Are omitted, and only the parts unique to the second embodiment will be described in detail.

[On the Encoding Side] On the encoding side, a synthesis filter information encoding unit 100, a weight filter 101, a formant emphasis information generation unit 102, a pitch component encoding unit 105, a noise component encoding unit 609, a gain component An encoding unit 110 and a local decoding unit 111 are provided. Among them, the pitch component encoding unit 105 includes the adaptive codebook 103 and the pitch drive signal search unit 104. The noise component coding section 609 includes a noise drive signal candidate generation section 606, a formant enhancement section 607, and a noise drive signal search section 608.

Next, the noise component coding unit 609 to which formant enhancement is applied based on the present invention will be described in detail. The noise drive signal candidate generation unit 606 is configured to generate a noise drive signal that can be uniquely decoded in association with a noise code by a predetermined method. The noise drive signal candidate generating section 606 can be normally realized by a statically provided noise codebook, but is not limited to this. A combination of a plurality of small noise codebooks or a combination in advance is limited. It may be one that expresses a static noise codebook corresponding to a pulse or the like, and the method of realizing the method is large. The present invention can cope with all the driving signal candidates generated in this way, and can partially or entirely deform the noise driving signal candidates so that encoding distortion can be dynamically suppressed by performing formant enhancement.

Here, the formant enhancement processing is performed only on some predetermined noise drive signal candidates, and the formant enhancement is not performed on the other noise drive signal candidates. An example of a method for realizing the configuration will be described. Noise drive signal candidate generator 6
06 is roughly divided into two parts, a first noise drive signal candidate generator (1) and a second noise drive signal candidate generator (2), and the noise drive signal candidate generated by the generator (1) is divided into two parts. , The formant enhancement is performed, and the noise drive signal candidates generated from the generation unit (2) are configured not to perform the formant enhancement.

The formant emphasis section 607 is configured to perform formant emphasis processing only on the drive signal candidates from the first noise drive signal candidate generation section (1).
Then, both the noise drive signal candidate subjected to the formant emphasis and the noise drive signal candidate not subjected to the formant emphasis are used, and a suitable noise drive signal having a smaller coding distortion in the noise drive signal search unit 608 and a noise corresponding thereto are obtained. Explore the code.

At this time, whether the noise drive signal finally selected is formant-enhanced is determined from the first noise drive signal generator (1) or the second noise drive signal generator ( It is clear that the judgment can be made based on whether the data comes from 2). In other words, if the correspondence of whether or not to formant-enhance the noise drive signal for each noise code is predetermined, the drive signal whose formant is enhanced by the finally selected noise code is used. It is easy to determine whether a drive signal without formant enhancement is used.

Also in this embodiment, in the same manner as the method described in the first embodiment, the search calculation is performed using a method in which the formant emphasis filter Hc (z) is included in the weighted synthesis filter in advance. It is clear that the amount can be reduced. At this time, the search in the case where the formant enhancement is not performed may be performed assuming that Hc (z) = 1.

[Decoding Side] Next, the decoding side will be described. 6, on the decoding side, a demultiplexing unit 113, a synthesis filter information reproducing unit 114, a formant emphasis information generating unit 115, an adaptive codebook 116, a noise drive signal reproducing unit 617, a formant emphasizing unit 619, a switching unit 620, A gain component reproduction unit 118, a drive signal generation unit 120, and a synthesis filter 121 are provided, and reproduce a synthesized audio signal based on the code separated by the demultiplexing unit 113. Hereinafter, each part will be described in detail.

The noise driving signal reproducing section 617 corresponds to the noise driving signal generating section 606 on the encoding side, and includes a first noise driving signal reproducing section (1) and a second noise driving signal reproducing section (2). A noise drive signal composed of two parts and corresponding to the noise code is reproduced in the same manner as on the encoding side. On the other hand, since it is known from the noise code whether or not the formant enhancement of the noise drive signal is performed, as described above, if the formant enhancement processing is necessary based on this information, the formant enhancement section 619 is used to formant the noise drive signal. Make emphasis. In the example of FIG. 6, the switching unit 620 is used to realize whether or not the formant enhancement of the drive signal on the decoding side is performed. However, the implementation method is not limited to this.

Next, the processing procedure on the encoding side in this embodiment will be described with reference to the flowchart shown in FIG. Normally, in the CELP method, encoding of synthesis filter information is performed in units of frames, and encoding of other pitch periods, noise drive signals, and gains is often performed in units of subframes obtained by further dividing a frame. For the sake of simplicity, an example will be described in which the pitch period, the noise drive signal, and the gain are designed to be encoded in frame units.

First, initial settings are made (step S30).
0), and then input a sound for frame processing (step S).
301), the information of the synthesis filter is extracted and encoded (step S302). Next, the input audio signal is weighted by a weight filter obtained from the analyzed information on the spectral envelope, and a target signal to be a target of the drive signal encoding is generated (step S303). Further, a search for a pitch drive signal is performed, and the selected pitch cycle is encoded (step S304). Then, using the information of the synthesis filter for formant enhancement, formant enhancement information that reflects the relationship between the unevenness of the spectrum envelope of the synthesis filter is obtained (step S305). Next, formant-emphasized noise drive signal candidates are generated for the candidates from the noise drive signal candidate generator (1) (step S).
315) Searching and encoding the noise drive signal using the formant-enhanced candidate and the formant-enhanced candidate so that the formant enhancement is not performed on the candidate from the noise drive signal generator (2). Is performed (step S306). Next, coding of gain values used for the pitch drive signal and the noise drive signal is performed (step S3).
07). The encoded data (encoded data) is multiplexed and transmitted to the decoding side as a bit stream (step S308).

Further, a drive signal is generated by giving a gain, and local decoding is performed to generate a synthesized speech signal with auditory weight (step S309). Finally, in preparation for the processing of the next frame, the internal data used for encoding is moved (step S310).

Next, the processing procedure on the decoding side in this embodiment will be described with reference to the flowchart shown in FIG. Here, an example using post-filter processing will be described. First, the internal state of the decoding process is initialized (step S400). Next, the bit stream (coded data) sent from the coding side is separated into codes used for decoding (step S401). Among these, the code of the synthesis filter is used to reproduce the synthesis filter information (step S402), and the pitch cycle code is used to reproduce the pitch drive signal (step S403). Next, formant emphasis information is obtained using the information of the synthesis filter (step S404).
Alternatively, a noise drive signal without formant enhancement is reproduced (step S405). Whether or not formant enhancement is performed can be determined by the value of the noise code. Next, the gain information is reproduced using the gain code (step S40).
6), the pitch driving signal and the step S
The noise driving signal reproduced in 405 is combined to obtain a driving signal (step S407). Then, this drive signal is input to a synthesis filter to create a synthesized voice (step S
408). This is passed through a post filter to generate a final synthesized speech for output (step S409). At the same time, the internal data is moved in preparation for the processing of the next frame (step S410).

(Third Embodiment) FIG. 9 is a block diagram showing the configuration of a speech encoding / decoding device according to a third embodiment. In the first and second embodiments, the configuration in which the formant enhancement is performed only on the noise drive signal candidates has been described. However, in the third embodiment, the formant enhancement is applied to both the pitch drive signal candidates and the noise drive signal candidates. An example of the configuration in the case of doing this will be described. In this embodiment,
Basically, except for the part related to pitch drive signal processing,
7, the same reference numerals as in FIG. 1 denote parts which can be realized by the same processing as in the first embodiment, and a description thereof will be omitted. The details related to the formant emphasis will be described in detail.

[Encoding side] In FIG. 9, on the encoding side, the synthesis filter information encoding section 100, the weight filter 101, the formant enhancement information generation section 702, the pitch component encoding section 705, and the noise component encoding section 709 are provided. , A gain component encoder 110 and a local decoder 111. Among them, the pitch component coding unit 705
Is the adaptive codebook 103 and the formant emphasis unit A7
03, a pitch drive signal search unit 104. The noise component coding unit 709 includes a noise drive signal candidate generation unit 106, a formant enhancement unit B707, and a noise drive signal search unit 108.

The formant emphasis information generation section 702 uses the information of the synthesis filter to obtain parameter information for formant emphasis that reflects the relationship between the unevenness of the spectral envelope of the synthesis filter. At this time, formant emphasis information A for the pitch drive signal and formant emphasis information B for the noise drive signal are output. By doing so, the degree and characteristics of formant emphasis can be optimized between the pitch drive signal and the noise drive signal, so that the sound quality can be further improved.

Next, the pitch component encoder 705 to which formant enhancement according to the present invention is applied will be described in detail. The adaptive codebook 103 stores past encoded drive signals, and generates pitch drive signal candidates for expressing pitch cycle components included in the target vector by a predetermined method.

The formant emphasis unit A 703 performs formant emphasis on the pitch drive signal candidate generated using the formant emphasis information A from the formant emphasis information generation unit 702. At this time, the pitch drive signal search unit 10
Reference numeral 4 searches for a candidate that reduces the distortion between the formant-emphasized weighted synthesized pitch signal and the target vector, which is defined using the formant-emphasized pitch drive signal candidate. The selected formant-emphasized pitch drive signal and the corresponding pitch period code,
Then, a weighted synthesized pitch signal with formant emphasis is output from the pitch component encoding unit 705.

Next, the noise component coding section 709 will be described. The noise drive signal candidate generating section 106 is configured to generate a noise drive signal that can be uniquely decoded in association with a noise code by a predetermined method. The noise drive signal candidate generating unit 106 can be generally realized by a statically provided noise codebook, but is not limited to this. A combination of a plurality of small noise codebooks or a combination in advance is limited. There is also known a method of expressing an equivalent to a static noise codebook by using a pulse or the like, and the method of realizing the method has a large degree of freedom. The present invention can cope with all such drive signal candidates, and can partially or entirely change the noise drive signal candidates so as to dynamically suppress coding distortion by performing formant enhancement.

The formant emphasis unit B 707 uses the formant emphasis information B from the formant emphasis information generation unit 702 to generate a formant-enhanced noise drive signal.

The noise drive signal searching section 108 combines the formant-enhanced noise drive signal candidates with a weighted synthesis filter, and uses the obtained synthesized noise signal to reduce distortion with respect to the target vector from which the pitch component has been removed. Such a candidate is searched for, and the selected noise code, the noise drive signal indicating the corresponding formant enhancement, and the weight-combined formant-enhanced noise signal are output as the output of the noise encoder 709.

[Decoding Side] In FIG. 9, the audio decoding unit includes a demultiplexing unit 113, a synthesis filter information reproducing unit 114, a formant emphasis information generating unit 715, an adaptive codebook 116, and a noise driving signal reproducing unit 117. , Formant emphasis section A719, formant emphasis section B720,
A gain component reproducing unit 118, a drive signal reproducing unit 120, and a synthesis filter 121 are provided, and reproduce a synthesized audio signal based on the code separated by the demultiplexing unit 113.

The formant emphasis information generation section 715 has the same function as the formant emphasis information generation section 702 described on the encoding side, and forms formant emphasis information A for a pitch drive signal and noise based on decoded synthesis filter information. It outputs formant emphasis information B for drive signal information.

The adaptive code book 116 stores past drive signals, and reproduces pitch drive signals based on pitch cycle codes. Next, the formant emphasis unit A719 performs formant emphasis on the reproduced pitch drive signal using the formant emphasis information A.

The noise driving signal reproducing section 117 reproduces the noise driving signal based on the noise code in the same manner as the encoding side.
The formant emphasis unit B720 performs formant emphasis on the reproduced noise drive signal using the formant emphasis information B.

Next, the processing procedure on the encoding side in this embodiment will be described with reference to the flowchart shown in FIG. Normally, in the CELP method, encoding of synthesis filter information is performed in units of frames, and encoding of other pitch periods, noise drive signals, and gains is often performed in units of subframes obtained by further dividing a frame. For the sake of simplicity, an example will be described in which the pitch period, the noise drive signal, and the gain are designed to be encoded in frame units.

First, initial settings are made (step S50).
0), and then input a sound for frame processing (step S).
501), the information of the synthesis filter is extracted and encoded (step S502). Next, the input audio signal is weighted by a weight filter obtained from the analyzed information on the spectrum envelope, and a target signal to be a target for encoding the drive signal is generated (step S503). Then, formant emphasis information A and formant emphasis information B that reflect the relationship between the unevenness of the spectrum envelope of the synthesis filter and the formant emphasis information B are obtained using the information of the synthesis filter for formant emphasis (step S504). The formant emphasis information A is for a pitch drive signal, and the formant emphasis information B is for a noise drive signal. Next, a pitch drive signal is searched for using the formant-emphasized information as a candidate for the formant-emphasized pitch drive signal,
Encode the selected pitch period (step S50)
5). Next, a noise drive signal search and encoding are performed using the noise drive signal candidates subjected to the formant enhancement using the formant enhancement information B as candidates (step S506). Next, the gain values used for the pitch drive signal and the noise drive signal are encoded (step S507). The coded data (coded data) is multiplexed and transmitted to the decoding side as a bit stream (step S50).
8).

Further, a drive signal is generated by giving a gain, and local decoding is performed to generate a synthesized speech with auditory weight (step S509). Finally, in preparation for the processing of the next frame, the internal data used for encoding is moved (step S510).

Next, the processing procedure on the decoding side in this embodiment will be described with reference to the flowchart shown in FIG.
First, the internal state of the decoding process is initialized (step S600). Next, the bit stream (coded data) sent from the coding side is separated into codes used for decoding (step S601). Of these, the synthesis filter information is reproduced using the synthesis filter code (step S602), and the formant enhancement information A and the formant enhancement information B are then generated using the synthesis filter information.
Is obtained (step S603).

Next, a pitch drive signal whose formant is emphasized based on the formant emphasis information A and the pitch cycle code is reproduced (step S604), and a noise drive signal whose formant is emphasized is reproduced based on the formant emphasis information B and the noise code. (Step S605). Next, the gain information is reproduced using the gain code (step S606), and the pitch drive signal subjected to formant emphasis and the noise drive signal subjected to formant emphasis are combined using the gain to obtain a drive signal (step S607). Next, the obtained drive signal is input to a synthesis filter to generate a synthesized voice (step S608). Finally, the internal data is moved in preparation for the next frame processing (step S609).

(Fourth Embodiment) FIG. 12 is a block diagram showing a configuration of a speech coding apparatus according to a fourth embodiment of the present invention. In this embodiment, another method capable of performing a search equivalent to the formant emphasis search of the noise drive signal in the encoding unit described in the first embodiment will be described. That is, the drive signal is encoded using the impulse response of the filter combining the auditory weighting synthesis filter and the formant enhancement filter (impulse response with formant enhancement) and the noise drive signal before formant enhancement. In the present embodiment, parts other than the noise component coding unit are the same as those of the first embodiment, and the description of the same parts will be omitted.

The encoding section includes a synthesis filter information encoding section 1
00, a weight filter 101, a formant enhancement information generator 102, a pitch component encoder 105, a pitch enhancement filter information generator 805, a noise component encoder 800, a gain component encoder 110, and a local decoder 111. You.

The noise component coding unit 800 among them
It comprises a noise drive signal candidate generator 804, an impulse response calculator with formant enhancement 801, a corrected target vector calculator 802, and a noise drive signal searcher 803. Hereinafter, each part will be described in detail.

The noise drive signal candidate generator 804 is configured to generate a noise drive signal that can be uniquely decoded by being associated with a noise code by a predetermined method. This noise drive signal candidate generating section 804 can be normally realized by a statically provided noise codebook, but is not limited to this. A combination of a plurality of small noise codebooks or a combination in advance is limited. It may be one that expresses a static noise codebook corresponding to a pulse or the like, and the method of realizing the method is large. The present invention can deal with all the drive signal candidates generated in this way.

The impulse response with formant emphasis calculation section 801 uses the information on the synthesis filter with auditory weights from the synthesis filter information encoding section 100 and the formant emphasis information from the formant emphasis information generation section 102 to generate the impulse response with formant emphasis. hwc (n) is calculated. Specifically, the characteristic H of the auditory weighted synthesis filter
w (z) and characteristic Hc (z) of the formant emphasis filter
Is used to obtain an impulse response hwc (n) of a hearing weighted synthesis filter with formant emphasis having a characteristic of Hwc (z) = Hw (z) Hc (z). This calculation is h
It can also be performed by wc (n) = hw (n) * hc (n). Here, hw (n) represents the impulse response of the synthesis filter with auditory weights, hc (n) represents the impulse response of the formant enhancement filter, and the symbol * represents calculation of convolution. It goes without saying that the impulse response of the M-order FIR filter using the above-described truncation by windowing can be used as hc (n).

The corrected target vector calculation unit 802 calculates a target vector (pitch component is preferably removed) x
Using formant enhancement with impulse response hwc (n) and calculates the corrected target vector x _m. Following an example of modified concrete calculation method of the target base vector x _m.

[0091]

(Equation 3)

Here, N indicates the number of samples in a section in which the noise drive signal is encoded. This calculation can be expressed as x _m = H _wc ^t x using matrix and vector. Here, H _wc represents an N × N lower triangular matrix representing a convolution operation using hwc (n) (the upper right element in the matrix is a lower triangular matrix of zero).

In addition to the above, not only the formant emphasis characteristic but also a method of emphasizing the pitch periodicity to the noise driving signal or a method of emphasizing the pitch periodicity to the impulse response of the perceptually weighted synthesis filter is provided. It is valid. Here, the latter method will be described. At this time, Hwc (z) can be expressed by Hwc (z) = Hw (z) Hc (z) Hp (z) (8) Further, the impulse response hwc (n) with formant emphasis can be obtained by hwc (n) = hw (n) * hc (n) * hp (n) (9) Here, Hp (z) and hp
(N) represents a pitch emphasis filter and its impulse response, respectively. An example of the transfer function of the pitch enhancement filter is represented by the following equation using the pitch period T and the pitch gain β supplied from the pitch enhancement filter information generation unit 805.

As the value of Hp (z) = 1 / (1-β- ^T ) T, the pitch period itself searched by the pitch component coding unit 105 or a pitch period determined based on the pitch period can be used. For example, when the pitch component _TF up to the precision of the non-integer sample is used in the pitch component encoding unit, the pitch period T = int (T _F ) is converted into an integer to reduce the amount of calculation of the pitch emphasis filter. The method used is also effective. It is clear that the order of the filters and the impulse responses in the equations (8) and (9) may be any order.

The noise drive signal search section 803 includes a corrected target vector x _m , an impulse response hwc (n) with formant enhancement, and a noise drive signal candidate from the noise drive signal candidate generation section 804 (a noise drive without formant enhancement). By using the signal candidate) vector v _k , a noise driving signal is selected by searching for a noise code k such that the following equation finally becomes as large as possible.

A _k ² / B _k = (x _m ^t v _k ) ² / (v _k ^t H _{w c} ^t H _{w c} v _k ) (11) In order to reduce the calculation amount required for searching for the noise drive signal,
Various methods are conceivable. For example, a large number of candidates are sifted to a small number of candidates using only the numerator part of expression (11), which is an evaluation expression for search, or the absolute value of x _m ^t v _k ,
If a method of selecting a drive signal candidate that maximizes the expression (11) from the remaining small number of candidates is used, the amount of calculation required for the search can be significantly reduced, and the quality hardly deteriorates compared to the full search. This is because, once x _m has been obtained before the search loop, the calculation of the numerator part of the equation (11) required in the search loop requires the corrected target vector x _m and the noise drive signal without formant emphasis. This is because it is only necessary to calculate the inner product directly with the candidate vector v _k, and the formant enhancement processing can be eliminated from the loop. In addition, as evaluation expressions for removing sieves, (x _m v _k ) ² f _k and | x
_m ^t v _k | f _k ^1/2 (where f _k = (v _k ^t v
_The use of _k ) ^-1 ) further improves the ability to sort out sieves, and has the effect of narrowing down to fewer candidates.

Further, the candidate of the noise drive signal is represented by a small number of pulses whose combination is limited in advance, and the absolute value of the pulse is restricted to a fixed value (for example, 1), or the absolute value amplitude of the pulse is Is in a narrow range (for example, 0.9 to
When the present invention is applied to a multi-pulse coding scheme having a special configuration constrained to 1.1), a drive signal candidate is not actually generated, and the element of the corrected target vector x _m is It is possible to search for the position where the absolute value is the maximum and narrow down the candidates for the position where the pulse rises, or to screen out the candidates. In particular, when the present invention is applied to a method using a drive signal candidate composed of only about 1 to 3 small pulses, a correction calculated using the target vector and the impulse response hwc (n) with formant emphasis. A search is made for a position where the absolute value of the element of the obtained target vector x _m is maximum, and the positive and negative of the pulse amplitude are determined in accordance with the positive and negative of the element of x _m at that position. It is also possible to determine the code. The multi-pulse method is described in detail in “Digital Voice Processing” (published by Tokai University Press, Furui) (Reference 2), and thus the description is omitted here.

When a noise code is selected, a formant-enhanced noise driving signal and a formant-weighted synthesized noise signal corresponding to the selected noise code are generated, and the noise code and these are output as outputs of the noise encoding unit. In the configuration in which the search is performed using not only the formant emphasis characteristic but also the pitch periodicity emphasis characteristic, both the formant emphasis and the pitch period from the pitch emphasis filter information generation unit 805 and the pitch periodicity emphasis using the pitch gain are emphasized. The generated noise driving signal is generated, and the weighted synthesized noise signal is generated by further performing auditory weighting synthesis.

Next, the encoding procedure in this embodiment will be described with reference to the flowchart shown in FIG. First, initial settings are made (step S800), and then a speech for frame processing is input (step S801), and information of a synthesis filter is extracted and encoded (step S802).

Next, the input audio signal is weighted by a weighting filter having characteristics obtained from the analyzed information on the spectral envelope, and a target signal to be a target of drive signal encoding is generated (step S803).

Further, a search for a pitch drive signal is performed, and the selected pitch cycle is encoded (step S804).
Then, using the information of the synthesis filter for formant enhancement, formant enhancement information that reflects the relationship between the unevenness of the spectral envelope of the synthesis filter is obtained.

When the pitch periodicity is emphasized, information on the pitch emphasis filter is further obtained (step S8).
05). Next, an impulse response with formant enhancement and a corrected target vector are calculated (step S80).
6), search and encode a noise drive signal using these (step S807).

Next, the gain values used for the pitch drive signal and the noise drive signal are encoded (step S80).
8). The coded data (coded data) is multiplexed and transmitted to the decoding side as a bit stream (step S809).

Further, a drive signal is generated by giving a gain, and local decoding is performed to generate a synthesized voice with auditory weight (step S810). Finally, the internal data used for encoding is moved in preparation for the processing of the next frame (step S811).

The embodiments of the present invention have been described above. However, the above-described embodiments are merely examples, and the formant enhancement of the drive signal according to the present invention is performed by analyzing the characteristics of the input voice and the method of analyzing the synthesis filter. It is needless to say that various application modifications are possible depending on. For example, the formant emphasis can be performed only on the pitch drive signal, or the judgment of not performing the formant emphasis may be made by analyzing a voice.

As auxiliary information of the formant emphasis information, information indicating the degree of formant emphasis and the presence or absence of formant emphasis may be encoded and transmitted from the encoding side to the decoding side.

In the present invention, the term "formant emphasis" is used for the sake of simplicity for speech. However, the application field is not limited to only speech such as a so-called telephone band. The present invention can also be applied to encoding of a voice, a tone signal, and the like.

Further, when a post filter is used on the decoding side in the present invention, a subjective improvement in sound quality is greater when a post filter having characteristics different from those of the post filter used in the CELP method or the like is used. In particular,
When the formant enhancement is applied to the drive signal according to the present invention, it is desirable that the formant enhancement performed inside the post-filter is adjusted to be weaker than usual. By doing so, the final post-filter output audio signal is appropriately and appropriately formant-emphasized, and the synthesized voice has the effect of having a more natural sound quality.

[0109]

As described above, according to the speech encoding / decoding method of the present invention, the distortion of the synthesized speech signal is improved by using a plurality of drive signal candidates including the formant-enhanced drive signal candidates. By selecting a drive signal to be reduced, encoding the selected drive signal, and generating a synthesized voice signal using the formant-enhanced drive signal on the decoding side as in the decoding side, for example, about 4 kbit / s. High quality synthesized speech can be generated even at the following low bit rates.

Further, by including a characteristic for correcting the inclination of the spectrum of the drive signal candidate in the formant emphasis characteristic for the drive signal candidate, an unnecessary spectrum inclination mixed in the formant emphasis processing is not given to the drive signal. This prevents a phenomenon in which the synthesized speech is muffled in time, and a more stable synthesized speech can be obtained.

Further, if the formant emphasis characteristic is obtained from information representing the characteristic of the synthesis filter, usually,
Since the information of the synthesis filter is transmitted from the encoding side to the decoding side, it is possible to transmit the information of the formant enhancement characteristic without transmitting special side information, that is, without increasing the amount of information, without increasing the amount of information. Quality can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech encoding / decoding device according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining a formant emphasis characteristic used in the present invention.

FIG. 3 is a diagram for explaining a formant emphasis characteristic used in the present invention.

FIG. 4 is a flowchart showing a procedure of a speech encoding process according to the first embodiment of the present invention;

FIG. 5 is a flowchart showing a speech decoding processing procedure according to the first embodiment of the present invention;

FIG. 6 is a block diagram showing a configuration of a speech encoding / decoding device according to a second embodiment of the present invention.

FIG. 7 is a flowchart showing a speech encoding processing procedure according to the second embodiment of the present invention;

FIG. 8 is a flowchart showing a speech decoding processing procedure according to the second embodiment of the present invention;

FIG. 9 is a block diagram illustrating a configuration of a speech encoding / decoding device according to a third embodiment of the present invention.

FIG. 10 is a flowchart showing a speech encoding processing procedure according to a third embodiment of the present invention.

FIG. 11 is a flowchart showing a speech decoding processing procedure according to the third embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration of a speech coding apparatus according to a fourth embodiment of the present invention.

FIG. 13 is a flowchart showing a speech encoding processing procedure according to the fourth embodiment of the present invention;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Speech signal input terminal 12 ... Speech signal output terminal 100 ... Synthesis filter information coding part 101 ... Weight filter 102,702 ... Formant emphasis information generation part 103 ... Adaptive codebook 104 ... Pitch drive signal search part 105,705 ... Pitch Component encoders 106, 606, 804 noise drive signal candidate generator 107, 607, 703, 707 formant enhancer 108, 608, 803 noise drive signal searcher 109, 800 noise component encoder 110 gain Component encoding unit 111 Local decoding unit 112 Multiplexing unit 113 Demultiplexing unit 114 Synthetic filter information reproducing unit 115, 715 Formant emphasis information generating unit 116 Adaptive codebook 117, 617 Noise driving signal reproduction Unit 118: gain component reproducing unit 119, 619, 719, 20 ... formant enhancement section 120 ... driving signal generation unit 121 ... synthesis filter 801 ... formant emphasis with the impulse response calculator 802 ... modified target vector calculation section 805 ... pitch emphasis filter information generation unit

Claims (14)

[Claims] A synthetic speech signal is expressed using a synthesis filter and a drive signal for driving the synthesis filter, and information and a drive signal representing characteristics of the synthesis filter for reducing distortion of the synthesized speech signal with respect to the input speech signal are reduced. In the voice coding method for coding, the drive signal is coded using the synthesis filter, a plurality of drive signal candidates, and a formant emphasis filter having characteristics obtained based on the input voice signal. Voice encoding method. 2. A method for expressing a synthesized speech signal using a synthesis filter and a drive signal for driving the synthesis filter, and information and a drive signal representing characteristics of the synthesis filter for reducing distortion of the synthesized speech signal with respect to the input speech signal. In the speech encoding method for encoding, using a plurality of drive signal candidates including a drive signal candidate formant-enhanced by a formant emphasis filter having characteristics obtained based on the input audio signal, the distortion of the synthesized audio signal is further reduced. A speech encoding method comprising: selecting a drive signal to be reduced; and encoding the selected drive signal. 3. A synthetic speech signal is expressed using a synthesis filter and a drive signal for driving the synthesis filter, and information and a drive signal representing characteristics of the synthesis filter for reducing distortion of the synthesized speech signal with respect to the input speech signal are obtained. In the speech encoding method for encoding, a formant emphasis characteristic is obtained from information representing characteristics of the synthesis filter, and the synthesized speech is obtained by using a plurality of drive signal candidates including a drive signal candidate formant-enhanced by the formant emphasis characteristic. A speech encoding method comprising: selecting a drive signal that reduces signal distortion; and encoding the selected drive signal. 4. A synthesized speech signal is expressed by using a synthesis filter and a drive signal composed of a combination of a pitch drive signal and a noise drive signal for driving the synthesis filter, thereby reducing distortion of the synthesized speech signal with respect to the input speech signal. In a speech encoding method for encoding information and a drive signal representing characteristics of a synthesis filter, at least a part of at least one of a pitch drive signal candidate and a noise drive signal candidate is formant by a formant enhancement characteristic obtained by a predetermined method. Emphasizing, using these pitch drive signal candidates and noise drive signal candidates, selecting a pitch drive signal and a noise drive signal that reduce distortion of the synthesized speech signal, and selecting the selected pitch drive signal and noise drive signal. A speech encoding method characterized by encoding a drive signal comprising a combination of the following. 5. The apparatus according to claim 1, wherein the formant emphasis characteristic includes a characteristic for correcting a spectrum inclination.
The speech encoding method according to any one of claims 1 to 4. 6. A method for reproducing information and a drive signal representing the characteristics of a synthesis filter from input coded data, and driving the synthesis filter whose characteristics are determined by the information representing the characteristics of the synthesis filter by a formant-enhanced drive signal. A speech decoding method characterized by obtaining a synthesized speech signal. 7. A synthetic filter whose characteristics are determined by the information representing the characteristics of the synthesis filter and the drive signal are reproduced from the input encoded data, and a characteristic of the synthesis filter is determined from the information of the synthesis filter. A speech decoding method characterized in that a synthesized speech signal is obtained by driving with a drive signal that is formant-enhanced with formant emphasis characteristics. 8. An information representing a characteristic of a synthesis filter that expresses a synthesized speech signal using a synthesis filter and a drive signal for driving the synthesis filter on the encoding side, and further reduces distortion of the synthesized speech signal with respect to an input speech signal. And transmitting encoded data obtained by encoding the drive signal, and reproducing a synthesized filter whose characteristics are determined by information representing characteristics of the synthesized filter reproduced from the encoded data on the decoding side from the encoded data. A speech encoding / decoding method for obtaining a synthesized speech signal by driving with a drive signal, wherein the encoding side has characteristics obtained based on the synthesis filter, a plurality of drive signal candidates, and the input speech signal. The drive signal is encoded using a formant emphasis filter, and the drive signal on the decoding side is formant-emphasized by the formant emphasis characteristic. Speech encoding / decoding method characterized by obtaining a synthesized speech signal by driving a more said synthesis filter. 9. An information representing a characteristic of a synthesis filter which expresses a synthesized speech signal using a synthesis filter and a drive signal for driving the synthesis filter on the encoding side, and further reduces distortion of the synthesized speech signal with respect to the input speech signal. And transmitting encoded data obtained by encoding the drive signal, and reproducing a synthesized filter whose characteristics are determined by information representing characteristics of the synthesized filter reproduced from the encoded data on the decoding side from the encoded data. A speech encoding / decoding method for obtaining a synthesized speech signal by driving with a drive signal, wherein a plurality of drive signal candidates including a formant-enhanced drive signal candidate with formant enhancement characteristics obtained on a coding side by a predetermined method Is used to select a drive signal that reduces the distortion of the synthesized speech signal, encodes the selected drive signal, and on the decoding side, Speech encoding / decoding method characterized by obtaining a synthesized speech signal by driving the synthesis filter by formant emphasized drive signal formant emphasis characteristic. 10. An information representing a characteristic of a synthesis filter that expresses a synthesized speech signal using a synthesis filter and a drive signal for driving the synthesis filter on the encoding side, and reduces distortion of the synthesized speech signal with respect to an input speech signal. And transmitting encoded data obtained by encoding the drive signal, and reproducing a synthesized filter whose characteristics are determined by information representing characteristics of the synthesized filter reproduced from the encoded data on the decoding side from the encoded data. A speech encoding / decoding method for obtaining a synthesized speech signal by driving with a drive signal, wherein a formant emphasis characteristic is obtained from information representing characteristics of the synthesis filter on the encoding side, and formant emphasis is performed using the formant emphasis characteristic. Using a plurality of drive signal candidates including drive signal candidates, select a drive signal to reduce the distortion of the synthesized audio signal, The selected drive signal is encoded. On the decoding side, a formant emphasis characteristic is obtained from information representing the characteristic of the synthesis filter reproduced from the encoded data. A speech encoding / decoding method characterized by obtaining a synthesized speech signal by driving a filter. 11. The encoding side formant-emphasizes at least a part of at least one of a pitch drive signal candidate and a noise drive signal candidate with a formant emphasis characteristic obtained by a predetermined method, and selects these pitch drive signal candidates. And a candidate for a noise drive signal, selecting a pitch drive signal and a noise drive signal that further reduce the distortion of the synthesized speech signal, and encoding a drive signal comprising a combination of the selected pitch drive signal and the noise drive signal. The speech encoding / decoding method according to claim 9 or 10, wherein: 12. The speech encoding / decoding method according to claim 8, wherein said formant emphasis characteristic includes a characteristic for correcting a spectrum inclination. 13. A synthetic voice signal is expressed by using a synthesis filter and a drive signal for driving the synthesis filter, and information and a drive signal representing characteristics of the synthesis filter for reducing distortion of the synthesized voice signal with respect to the input voice signal are reduced. In the voice coding method to be coded, the characteristics of a perceptual weighting synthesis filter and the characteristics of a formant emphasis filter and a target vector used for coding the drive signal are determined based on the input voice signal, and the perceptual weighting synthesis filter and Determining a formant-enhanced impulse response using a formant emphasis filter; obtaining the target vector and a corrected target vector using the formant-enhanced impulse response; Driven with Impulse Response A speech encoding method comprising encoding a signal. 14. A synthetic speech signal is represented by using a synthesis filter and a drive signal for driving the synthesis filter, and information and a drive signal representing characteristics of the synthesis filter for reducing distortion of the synthesized speech signal with respect to the input speech signal are reduced. In the voice coding method to be coded, a characteristic and a target vector of a perceptual weighting synthesis filter, a pitch emphasis filter, and a formant emphasis filter used for coding the drive signal are obtained based on the input voice signal, and the perceptual weighting synthesis is performed. A filter, a pitch emphasis filter and a formant emphasis filter are used to obtain an impulse response with formant emphasis, the target vector and the corrected target vector are obtained using the formant emphasis impulse response, With target vector and formant enhancement A speech encoding method comprising encoding a noise drive signal using an impulse response.