JP2001100799A - Method and device for sound encoding and computer readable recording medium stored with sound encoding algorithm - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the tone quality by reducing the difference between the pitch period of an adaptive code vector and that of a sound source code vector without increasing the operation volume required for search of a sound source code book. SOLUTION: A digital sound input signal is divided into frames by a frame division part 12, and they are divided into sub-frames by a sub-frame division part 14, and a formant parameter extraction part 16 analyzes frames to extract and encode formant parameters, and a pitch period extraction part 18 uses formant parameters of sub-frames and frames to extract and encode the pitch period of sub-frames, and a noise source extraction part 20 extracts and encodes noise source components of sub-frames; and when a pulse will be made rise in a non-integer sample position, pulses are made rise in plural integer sample positions between which the non-integer sample position is interposed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a speech coding apparatus,
For voice encoding method and recording medium, more specifically, for example, answering machine, facsimile machine with answering machine,
Voice coding apparatus, voice coding method and voice coding algorithm for coding a digital voice signal used in voice coding application systems such as voice communication, voice memo, voice mail system, digital cellular phone, DSVD modem, or Internet phone The present invention relates to a computer-readable recording medium having recorded thereon.

[0002]

2. Description of the Related Art In recent years, mobile phones, Internet phones,
With the widespread use of applications such as DSVD modems that use voice coding technology for coding voice signals at a low bit rate, there is an increasing demand for higher quality voice coding technology.

[0003] For example, the mainstream of speech coding at 6 kbps to 16 kbps is CELP (Code Excited L).
inear Prediction coding system)
A system using the concept of LP is used in various applications described above and various standards such as digital mobile phones in Japan.

The CELP realizes low bit rate coding by quantizing vocal tract information and vector-quantizing the vocal tract information using a speech source / vocal tract separation model based on linear prediction. . Specifically, first, the input voice is divided into frames, which are processing units, and the frames are analyzed to extract and quantize formant parameters representing the formant components of the voice corresponding to the vocal tract information.
The quantization of the excitation information is often performed on a subframe shorter than the frame, and is often performed in two stages using two types of codebooks.

[0005] The first stage is an adaptive codebook search using an adaptive codebook representing pitch information of speech. The adaptive codebook generates a vector (adaptive code vector) having a subframe length by repeating the excitation information up to the immediately preceding subframe by the pitch length for a predetermined number of pitches.
It is configured adaptively for each subframe. Next, each adaptive code vector is passed through a vocal tract filter constituted by the formant parameters quantized as described above to generate a synthesized vector. Further, the distance between the combined vector and the input subframe is measured, and the optimal adaptive code vector, that is, the optimal pitch is determined to give the shortest distance.

The second stage is called an excitation codebook (also called a noise codebook). An excitation codebook search (also referred to as a noise codebook search) using a fixed codebook of a predetermined number of vectors (excitation code vectors). The types of the excitation codebook include a learning codebook configured by learning, a random codebook configured by random numbers, an algebraic excitation codebook configured algebraically using several pulses, or a partial There was a combination.

In this case, first, a synthesized vector obtained by passing the determined optimal adaptive code vector through a vocal tract filter is subtracted from the subframe vector,
A target signal for excitation codebook search is generated. Next, a combined vector is generated from each excitation code vector through the vocal tract filter in the same manner as the adaptive code vector. Further, the distance between the target signal of the excitation codebook search and the composite vector is measured, and the optimal excitation code vector that gives the shortest distance is determined.

[0008]

However, in the above-mentioned conventional speech coding apparatus and method, the excitation codebook is for expressing components other than the formant component and the pitch component of the speech. If the pitch length is shorter than the subframe, the target signal of the excitation code search will have a periodicity based on the pitch. For this reason, a problem arises that distortion is increased in the ordinary excitation codebook search.

[0009] Naturally, this property becomes remarkable as the subframe length becomes longer. In recent years, with the spread of Internet telephones, there has been a demand for higher quality encoding at a low delay bit rate near 4 kbps. In this case, in order to realize a low-delay scheme at 4 kbps using CELP, the subframe length must be 5 to 5 which has been conventionally used.
Since it is necessary to set the time to about 10 ms, which is longer than about 8 ms, the necessity of solving the above problem becomes more important.

Therefore, as one of the methods for addressing this problem, the excitation code vector is made periodic in correspondence with the pitch period, so that a periodic but non-stationary component can be expressed. A pitch periodization method for improving the coding quality of voiced speech has been proposed, and this method is still used at present.

There are two types of pitches, where one sample length is an integer unit, an integer length pitch (integer pitch) and a decimal length pitch (non-integer pitch). By using the method, more accurate pitch extraction becomes possible. The adaptive code vector corresponding to this non-integer pitch is disclosed in Japanese Patent Application Laid-Open No. 9-319396 and the paper "Miki, Moriya, Mano, Omuro," CELP coding with pitch synchronous noise excitation source (PSI-CELP) ". IEICE (A), Vol.J77-A, No.3, pp.3
14-324 (1994) "and the like, and can be configured by upsampling using the sampling theorem.

In the algebraic excitation codebook, the excitation codevector is constituted by a plurality of pulses. In the case of a codebook having the same size (the same number of excitation codevectors), the codebook differs from other codebooks. It is widely used because it has a characteristic that the amount of memory required for storing a codebook and the amount of computation for searching are very small.

However, when the pulses constituting the algebraic excitation code vector are subjected to periodicity corresponding to a non-integer pitch using the sampling theorem, it is necessary to raise pulses at many sample points, and the multipulse excitation The low computational complexity of the is lost. So, in this case,
A method of approximately performing pitch periodization by using an integer part obtained by truncating a part below a decimal point is used.

However, this technique can easily and easily realize pitch pitching, but has a problem that the difference affects the sound quality because the cycle of the adaptive code vector differs from the cycle of the noise source vector. Was. There is a problem that the influence on the sound quality is particularly large in the case of a short pitch such as a female sound.

The present invention has been made in view of the above, and reduces the difference between the pitch period of an adaptive code vector and the pitch period of an excitation code vector without greatly increasing the amount of computation required for excitation codebook search. It is an object of the present invention to provide a computer-readable recording medium that describes a voice coding device, a voice coding method, and a voice coding algorithm that can improve the sound quality by performing the above.

[0016]

According to one aspect of the present invention, there is provided an audio encoding apparatus for encoding a digital audio signal, wherein the digital audio signal is divided into frames. Frame dividing means, sub-frame dividing means for dividing the frame into shorter sub-frames, formant parameter extracting means for extracting and encoding formant parameters from the frame,
Pitch period extracting means for extracting and encoding a pitch period of the subframe using the subframe and the formant parameter, and a subframe constituted by the subframe, the formant parameter, the pitch period, and a plurality of pulses. Noise source extraction means for extracting and encoding a noise source component of the sub-frame using a multi-pulse codebook constituted by a plurality of long code vectors, wherein the noise source extraction means If you need to pulse a position,
It is characterized in that a pulse is raised at a plurality of integer sample positions sandwiching the non-integer sample position. According to the first aspect of the present invention, noise source extraction can be performed with high accuracy, and higher sound quality can be achieved.

Further, the invention described in claim 2 is the first invention.
In the speech encoding apparatus described above, when the noise source extracting unit needs to generate a pulse at a non-integer sample position α, when the fractional part of α is β (0 <β <1), α
A pulse having an amplitude of (1−β) times the pulse amplitude is set up at the smaller maximum integer sample position (α−β), and the pulse of β times the pulse amplitude is set at the minimum integer sample position (α−β + 1) larger than α. It is characterized in that a pulse having an amplitude is set up. According to the second aspect of the present invention, noise source extraction can be performed with a simpler configuration with high accuracy and high sound quality can be achieved.

According to a third aspect of the present invention, there is provided an audio encoding apparatus for encoding a digital audio signal, comprising: frame dividing means for dividing the digital audio signal into frames; Subframe dividing means for dividing, formant parameter extracting means for extracting and encoding formant parameters from the frame, and pitch period for extracting and encoding the pitch period of the subframe using the subframe and formant parameters Extracting means, the sub-frame,
Noise that extracts and encodes a noise source component of the subframe using a multi-pulse codebook composed of a plurality of code vectors having a subframe length composed of the formant parameter, the pitch period, and a plurality of pulses. Source extraction means, and further, the pitch cycle extraction means extracts not only an integer pitch cycle but also a non-integer pitch cycle, and the noise source extraction means converts the code vector corresponding to the pitch cycle. Providing a pitch periodicizing means for periodicizing, and when it is necessary for the pitch periodicizing means to generate a pulse at a non-integer sample position, generating a pulse at a plurality of integer sample positions sandwiching the non-integer sample position It is characterized by. According to the third aspect of the present invention, the pitch period of the multi-pulse type sound source with respect to the non-integer pitch can be made with high accuracy, and the sound quality can be further improved.

The invention described in claim 4 is the same as the invention described in claim 3.
In the speech coding apparatus described above, when it is necessary to set a pulse at the non-integer sample position α by the pitch periodizing means, when the fractional part of α is β (0 <β <1),
A pulse having an amplitude of (1−β) times the pulse amplitude is set at the largest integer sample position (α−β) smaller than α, and β times the pulse amplitude at the smallest integer sample position (α−β + 1) larger than α. A pulse having an amplitude of? According to the fourth aspect of the present invention, the pitch period of the non-integer pitch of the multi-pulse type sound source can be highly accurately performed with a simpler configuration, and the sound quality can be improved.

According to a fifth aspect of the present invention, there is provided a voice coding method for coding a digital voice signal, wherein the digital voice signal is divided into frames, and the frame is divided into shorter subframes. A subframe dividing step of dividing, a formant parameter extracting step of extracting and encoding a formant parameter from the frame, and a pitch period of extracting and encoding a pitch period of the subframe using the subframe and the formant parameter An extraction step, the sub-frame,
Noise that extracts and encodes a noise source component of the subframe using a multi-pulse codebook composed of a plurality of code vectors having a subframe length composed of the formant parameter, the pitch period, and a plurality of pulses. And a source extraction step, wherein in the noise source extraction step, when a pulse needs to be made at a non-integer sample position, a pulse is made at a plurality of integer sample positions sandwiching the non-integer sample position. Features. According to the fifth aspect of the invention, noise source extraction can be performed with high accuracy, and higher sound quality can be achieved.

The invention described in claim 6 is the same as the claim 5
In the speech coding method described above, when it is necessary to make a pulse at a non-integer sample position α in the noise source extraction step, when the fractional part of α is β (0 <β <1), α
A pulse having an amplitude of (1−β) times the pulse amplitude is set up at the smaller maximum integer sample position (α−β), and the pulse of β times the pulse amplitude is set at the minimum integer sample position (α−β + 1) larger than α. It is characterized in that a pulse having an amplitude is set up. According to the sixth aspect of the present invention, noise source extraction can be performed with a simpler configuration with high accuracy and high sound quality can be achieved.

According to a seventh aspect of the present invention, there is provided a voice coding method for coding a digital voice signal, comprising: a frame dividing step of dividing the digital voice signal into frames; A subframe dividing step of dividing, a formant parameter extracting step of extracting and encoding a formant parameter from the frame, and a pitch period of extracting and encoding a pitch period of the subframe using the subframe and the formant parameter An extraction step, the sub-frame,
Noise that extracts and encodes a noise source component of the subframe using a multi-pulse codebook composed of a plurality of code vectors having a subframe length composed of the formant parameter, the pitch period, and a plurality of pulses. Source extraction step, and further, the pitch period extraction step extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extraction step converts the code vector corresponding to the pitch period. Including a pitch cycling step of cycling, when the pitch cycling step needs to raise a pulse at a non-integer sample position, raise a pulse at a plurality of integer sample positions sandwiching the non-integer sample position It is characterized by the following. According to the seventh aspect of the present invention, the pitch period of the non-integer pitch of the multi-pulse type sound source can be made with high accuracy, and the sound quality can be further improved.

The invention described in claim 8 is the same as the invention described in claim 7.
In the speech encoding method described above, when it is necessary to make a pulse at a non-integer sample position α in the pitch periodization step, when the fractional part of α is β (0 <β <1),
A pulse having an amplitude of (1−β) times the pulse amplitude is set at the largest integer sample position (α−β) smaller than α, and β times the pulse amplitude at the smallest integer sample position (α−β + 1) larger than α. A pulse having an amplitude of? According to the eighth aspect of the present invention, the pitch period of the non-integer pitch of the multi-pulse type sound source can be highly accurately performed with a simpler configuration, and the sound quality can be improved.

According to a ninth aspect of the present invention, there is provided a computer-readable recording medium recording an audio encoding algorithm for encoding a digital audio signal, comprising: a frame dividing step for dividing the digital audio signal into frames. Sub-frame division step of dividing the frame into shorter sub-frames, formant parameter extraction step of extracting and encoding formant parameters from the frame, and pitch cycle of the sub-frame using the sub-frame and formant parameters. Pitch period extracting step of extracting and encoding
The sub-frame, the formant parameters, the pitch period and a noise source component of the sub-frame is extracted using a multi-pulse codebook composed of a plurality of code vectors of a sub-frame length configured by a plurality of pulses. A noise source extracting step of encoding, and further comprising the step of extracting not only an integer pitch period but also a non-integer pitch period, and the noise source extracting step converts the code vector into the pitch period. Including a pitch periodicization step that is periodicized according to
The pitch period setting step is characterized in that, when a pulse needs to be raised at a non-integer sample position, a pulse is raised at a plurality of integer sample positions sandwiching the non-integer sample position.

According to the ninth aspect of the present invention, a speech encoding algorithm capable of performing a pitch cycle with respect to a non-integer pitch of a multi-pulse type sound source with high precision and achieving higher sound quality is recorded on a recording medium. By recording, the speech encoding algorithm becomes machine-readable, and these operations can be realized by a computer.

According to a tenth aspect of the present invention, in the computer readable recording medium having recorded thereon the speech encoding algorithm according to the ninth aspect, it is necessary to form a pulse at a non-integer sample position α in the pitch periodizing step. Occurs, and when the fractional part of α is β (0 <β <1), the largest integer sample position smaller than α (α−
β), a pulse having an amplitude of (1−β) times the pulse amplitude is set, and the smallest integer sample position (α−β +
In 1), a pulse having an amplitude β times the pulse amplitude is set.

According to the tenth aspect of the present invention, there is provided a recording medium for a speech encoding algorithm capable of performing a pitch period with respect to a non-integer pitch of a multi-pulse type sound source with high accuracy and a high sound quality with a simpler configuration. By recording in
The speech encoding algorithm becomes machine-readable, and these operations can be realized by a computer.

[0028]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration example of a speech encoding device according to the present embodiment. In this example, CELP (Code Excited Linear Prediction codin) is used.
g system) for encoding a digital audio input signal.

The speech encoding device 10 includes a frame dividing unit 1
2, subframe division unit 14, formant parameter extraction unit 16, pitch period extraction unit 18, noise source extraction unit 20,
And a gain quantization unit 30. Further, the above-described noise source extraction unit 20 is configured by a target signal configuration unit 22, a pitch periodization unit 24, a distance calculation unit 26, an optimum vector determination unit 28, and the like.

The frame dividing section 12 divides the digital audio input signal into processing units called frames. The frame length has a value of, for example, 20 to 30 ms, and a DC component may be removed by passing a high-pass filter before and after frame division.

The sub-frame division unit 14 divides a frame into processing units called sub-frames. The number of sub-frames per frame is, for example, 2 to 6.

The formant parameter extracting section 16 analyzes the divided frames and extracts formant parameters. As the formant parameter, L
PC (Linear Prediction Coding) coefficient, L
SP (Line Spectrum Frequency), LPC spectrum coefficient, reflection coefficient, and the like. The method of extracting formant parameters includes, for example,
There is linear prediction analysis and the like. As a method of the linear prediction analysis, an autocorrelation function of a frame is calculated, and Levinson-Dur
There is a method of calculating LPC coefficients by a recursive solution of bin, and a window function such as a Hamming window or a Hanning window may be applied to a frame before calculating an autocorrelation function.

The formant parameters of the extracted frame are quantized and transmitted or stored. As a method of quantizing the formant parameter of a frame, there are scalar quantization, vector quantization, multi-stage vector quantization, split vector quantization, and the like. When quantizing the formant parameter, it is desirable to use a parameter with high quantization efficiency such as LSP or LSF.

The formant parameter of the sub-frame is calculated using the quantized formant parameter of the frame. As a method of calculating the formant parameters of the subframe, there is a method of obtaining the formant parameters of the current and past quantized frames by interpolation, and examples of the interpolation methods include linear interpolation and quadratic interpolation.

The pitch period extracting section 18 extracts a pitch period for each of the above-mentioned subframes.
Examples of the pitch period extraction method include a method based on an adaptive codebook search and a method using an autocorrelation function. For example, in the case of an adaptive codebook search, a predetermined number of pitch candidates are prepared in advance. The pitch candidates include an integer pitch that is an integral multiple of the sampling unit and a non-integer pitch.
The following processing is performed on all of the pitch candidates.

First, an adaptive code vector is generated. This adaptive code vector is generated by cutting out the pitch length from the excitation vector up to the immediately preceding subframe and arranging it repeatedly until the subframe length is reached. Then, a synthesized speech vector is generated by using the adaptive code vector and the formant parameter of the subframe. As a method of generating the synthesized speech vector, there is a method of applying a linear filter configured by the formant parameters of the subframe to the adaptive code vector.

When the synthesized speech vector is multiplied by an arbitrary gain, the shortest distance closest to the subframe is calculated. In the calculation of the shortest distance, a method may be introduced in which auditory weighting is performed to minimize an auditory error. The above processing is performed on all the pitches in the pitch candidate, and the adaptive code vector having the shortest distance and the minimum pitch are determined. The code assigned to the pitch period of the subframe is transmitted or stored.

The noise source extraction unit 20 extracts and encodes a noise source component. An excitation codebook search using a multipulse excitation codebook composed of a plurality of excitation code vectors having a subframe length composed of a plurality of pulses is used as a method of extracting the noise source component.

The target signal forming section 22 generates a target signal vector for noise source search by subtracting from the subframe a synthesized speech vector synthesized by the adaptive code vector of the subframe and the formant parameter of the subframe. I do.

The pitch period generator 24 performs pitch period adaptation for each excitation code vector in accordance with the extracted pitch period of the subframe. In this pitch period technique, the subframe length is M, the pitch period of the subframe is α, the pulse position is P (0 ≦ P <M), and the pitch period α is an integer, It can be non-integer. In the case of P <α <M, as shown in FIG. 2, the pitch period of the pulse is performed. For example, when α is an integer, P + nα (n is a positive integer, provided that P + nα
A pulse is raised at the position <M).

When α is a non-integer, P + nα may become a non-integer, and it is necessary to raise a pulse at a non-integer sample position. At least when n = 1, it is a non-integer. Therefore, a plurality of pulses are generated by approximating the pulse to a triangular wave and interpolating, and the pulses at the non-integer sample positions are substituted by the plurality of pulses (see FIG. 3).

For example, taking n = 1 as an example, consider a triangular waveform having an apex at P + α and a base length of 2 samples. The part below the decimal point of the pitch period α is β (0 <β <
Assuming that 1), for two integer sample positions sandwiching P + α, a pulse having a height of (1−β) times is generated at the position of P + α−β, and β is set at the position of P + α−β + 1.
Make a pulse twice as high. Thus, by approximating the pulse to a triangular waveform and interpolating,
Even when it is necessary to make a pulse at a non-integer sample position, it is possible to substitute a small number of pulses approximately (see FIG. 4).

The distance calculator 26 performs the same distance calculation as in the adaptive codebook search for each pitch-periodic multi-pulse excitation code vector, and the synthesized speech vector of each excitation code vector is multiplied by an arbitrary gain. In this case, the shortest distance closest to the target signal vector is calculated. In this case, the distance calculation may be performed after making the synthesized speech vector of the excitation code vector orthogonal to the synthesized speech vector of the adaptive code vector.

The optimum vector determination unit 28 determines the excitation code vector that minimizes the shortest distance. The code assigned to the determined excitation code vector is transmitted or stored.

The gain quantization section 30 quantizes a gain component. As a method of quantizing the gain component, there are scalar quantization for separately quantizing the gain component of the adaptive code vector and the gain component of the noise code vector, and vector quantization for quantizing both of them at the same time. Is raised. The quantized gain component is transmitted or stored.

The operation of the above configuration will be described. FIGS. 5 and 6 are flowcharts each showing an operation example of the encoding processing of the digital audio input signal. Common processing operations will be described together. First, the frame division unit 12 divides a digital audio input signal into processing units called frames, and removes a DC component by passing a high-pass filter after (or before) the frame division (steps S1 and S2).
1).

Then, the frame divided by the sub-frame dividing section 14 is further divided into processing units called sub-frames (steps S2 and S22). Then, step S
In 3 (S23), i = 0 (i is a counter for checking the number of subframes), and in step S4 (S24), a formant parameter extraction process is performed. That is, the formant parameter extraction unit 16 analyzes the divided frames and extracts the formant parameters. As a method of extracting the formant parameters, a linear prediction analysis technique of calculating an autocorrelation function of a frame and calculating an LPC (linear prediction) coefficient by a Levinson-Durbin recursive method is used.

Next, the formant parameters of the frame are quantized and transmitted or stored. The formant parameter quantization method includes scalar quantization, vector quantization, multi-stage vector quantization, split vector quantization, and the like. The formant parameter is quantized using a parameter with high quantization efficiency such as LSP or LSF.

Next, the formant parameter of the sub-frame is calculated using the quantized formant parameter of the frame. This subframe formant parameter calculation method can be obtained by interpolation from the formant parameters of the current and past quantized frames, and linear interpolation or quadratic interpolation is used as the interpolation method.

Next, the pitch cycle extracting section 18 extracts a pitch cycle from each of the above-described subframes (steps S5 and S25). Here, an adaptive codebook search is used as a pitch period extraction method. In the adaptive codebook search, a predetermined number of pitch candidates including an integer pitch that is an integral multiple of the sampling unit and a non-integer pitch are prepared in advance, and the following processing is performed on all of the pitch candidates.

First, an adaptive code vector is generated by cutting out the pitch length from the excitation vector up to the immediately preceding subframe and arranging it repeatedly until the subframe length is reached. Then, a synthesized speech vector is generated by applying a linear filter formed by the formant parameters of the subframe to the adaptive code vector. Further, when the synthesized speech vector is multiplied by an arbitrary gain, the shortest distance closest to the subframe is calculated. Here, in calculating the shortest distance, an auditory weighting is performed so that an auditory error is minimized.

Such processing is performed for all the pitches in the pitch candidate, and the code assigned to the pitch cycle of the subframe is transmitted or stored as the adaptive code vector having the shortest distance and the pitch cycle. .

Next, the noise source extraction unit 20 extracts and encodes a noise source component. An excitation codebook search using a multipulse excitation codebook composed of a plurality of excitation code vectors having a subframe length composed of a plurality of pulses is used as a method of extracting a noise source component.

First, in the target signal configuration processing in step S6 (S26), the target signal configuration unit 22 converts the adaptive speech vector synthesized by the above-described adaptive code vector of the subframe and the formant parameter of the subframe into the subframe. , A target signal vector for noise source search is generated. And step S
At 7 (S27), j = 0 (j is a pulse number check counter), and the routine goes to Step S8 (S28).

In the pitch period generator 24, the respective excitation code vectors are subjected to the pitch period adaptation to the extracted pitch period of the subframe. here,
The subframe length (number) is M, the pitch period of the subframe is α, and the pulse position is P (0 ≦ P <M). The pitch period α is either an integer or a non-integer. possible.

In FIG. 5, in step S8, it is determined whether or not the position is a non-integer sample position. If the position is a non-integer sample position, one pulse is raised (step S9). A pulse is set up (step S10).

In FIG. 6, when P <α <M, the pitch of the pulse is cycled (step S28). When α is an integer, P + nα (n is a positive integer, provided that P + n
A pulse is generated at the position of α <M (see FIG. 2). α
Is a non-integer, P + nα may be a non-integer, that is, a pulse needs to be raised at a non-integer sample position (step S29). At least, when n = 1, it is a non-integer.

Here, a plurality of pulses are generated by approximating the pulse to a triangular wave and performing interpolation, and a pulse at a non-integer sample position is substituted by the plurality of pulses (see FIG. 3). For example, if n = 1, P +
Consider a triangular waveform with an apex angle at α and a base length of 2 samples. The part below the decimal point of the pitch period α is β (0 <β
Assuming that <1), for two integer sample positions sandwiching P + α, a pulse having a height of (1−β) times is set at the position of P + α−β (step S30), and β is set at the position of P + α−β + 1. A pulse having a double height is set up (step S31).

As described above, when it is necessary to raise a pulse at a non-integer sample position by approximating the pulse to a triangular waveform and performing interpolation (steps S28 to S29).
However, it is possible to substitute a small number of pulses approximately (see FIG. 4).

Next, distance calculating section 26 performs the same distance calculating process (step S) as described above for the adaptive codebook search for each pitch-periodic multipulse excitation code vector.
11, S32) is performed, and the shortest distance at which the synthesized speech vector of each excitation code vector is closest to the target signal vector is calculated by multiplying the gain by an arbitrary gain. In this case, the distance can be calculated after orthogonalizing the synthesized speech vector of the excitation code vector with respect to the synthesized speech vector of the adaptive code vector.

Next, in the optimum vector determination unit 28, the excitation code vector whose shortest distance is the smallest is determined, and the code assigned to the determined excitation code vector is transmitted or stored.

Further, the gain component is quantized in the gain quantization section 30. Here, as a gain component extraction method, vector quantization for simultaneously optimizing and quantizing both gain components of the adaptive code vector and the noise code vector is used, and the quantized gain component is transmitted or stored. (Steps S12 and S33).

Then, in step S13 (S34), it is determined whether or not j = N-1 (N is the number of pulses).
If j ≠ N-1, j = j + 1 is set in step S14 (S35), and the process returns to step S8 (S28).
If = N-1, the process moves to step S15 (S36).

In step S15 (S36), i ≠
If M−1 (M is the number of subframes), step S
16 (S37), i = i + 1 is set, and step S5 is performed.
Returning to (S25), if i = M-1, step S1
7 (S38). In step S17 (S38), if the processing frame is not the last frame, the process returns to the first step S1 (S21) and the above processing is repeated.

As described above, according to the above-described embodiment, when it is necessary to make a pulse at a non-integer sample position in the pitch period of CELP coding using an algebraic excitation codebook, Is approximated by a triangular wave, and is converted into a combination of pulses at integer sample positions by interpolation. Therefore, the pitch period of the adaptive code vector and the pitch period of the excitation code vector can be increased without greatly increasing the amount of computation required for excitation codebook search. And the sound quality can be improved.

The speech coding method described in the above embodiment is realized by executing a prepared speech coding algorithm (or a program including the same) on a computer such as a personal computer or a workstation. be able to. This audio encoding algorithm is recorded on a computer-readable recording medium such as a hard disk, a floppy disk, a CD-ROM, an MO, and a DVD, and is executed by being read from the recording medium by the computer.
Also, the audio encoding algorithm can be distributed via the recording medium and a network such as the Internet.

[0067]

As described above, according to the first aspect of the present invention, the frame dividing means divides the digital audio signal into frames, and the sub-frame dividing means divides the frame into shorter sub-frames. When the parameter extracting unit extracts and encodes the formant parameter from the frame, the pitch period extracting unit extracts and encodes the pitch period of the subframe using the subframe and the formant parameter, and the subframe and the noise source extracting unit encode the pitch period. A noise source component of a subframe is extracted and encoded using a formant parameter of the frame, a pitch period, and a multi-pulse codebook composed of a plurality of code vectors having a subframe length composed of a plurality of pulses. When the noise source extraction means needs to generate a pulse at a non-integer sample position, a pulse is generated at a plurality of integer sample positions sandwiching the non-integer sample position. A speech encoding device that can be performed with high accuracy and that can achieve higher sound quality can be obtained.

According to the second aspect of the present invention, the first aspect is provided.
In the invention of (1), when it is necessary to raise a pulse at the non-integer sample position α by the noise source extracting means, when the fractional part of α is β (0 <β <1), the largest integer sample position ( A pulse having an amplitude of (1-β) times the pulse amplitude is set at (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. Therefore, a speech coding apparatus capable of performing noise source extraction with high accuracy and a high sound quality with a simpler configuration can be obtained.

According to the third aspect of the present invention, the frame dividing means divides the digital audio signal into frames, the subframe dividing means divides the frame into shorter subframes, and the formant parameter extracting means divides the frame into shorter subframes. When the formant parameter is extracted and encoded, the pitch period extracting means extracts and encodes the pitch period of the subframe using the subframe and the formant parameter, and the noise source extracting means encodes the formant parameter and the pitch of the subframe and the frame. A noise source component of a sub-frame is extracted and encoded using a period and a multi-pulse codebook composed of a plurality of code vectors having a sub-frame length composed of a plurality of pulses. The pitch period extracting means extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extracting means includes a pitch periodizing means for periodicizing a code vector corresponding to the pitch period. When it is necessary for the periodicization means to make a pulse at a non-integer sample position, a pulse is made at a plurality of integer sample positions sandwiching the non-integer sample position. Can be performed with high precision, and a speech encoding device that can achieve higher sound quality can be obtained.

Further, according to the invention of claim 4, according to claim 3,
In the invention of (1), when it is necessary to raise a pulse at the non-integer sample position α by the pitch periodicizing means, when the fractional part of α is β (0 <β <1), the largest integer sample position (α) A pulse having an amplitude of (1-β) times the pulse amplitude is set at (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. Therefore, it is possible to obtain a speech encoding device that can perform a pitch period with respect to a non-integer pitch of a multi-pulse type sound source with high accuracy with a simpler configuration and achieve high sound quality.

According to the invention of claim 5, the digital audio signal is divided into frames in the frame dividing step, the frame is further divided into shorter sub-frames in the sub-frame dividing step, and the frame is extracted from the frame in the formant parameter extracting step. The formant parameters are extracted and encoded, the pitch period of the subframe is extracted and encoded by using the subframe and the formant parameter in the pitch period extracting step, and the formant parameter and the pitch period of the subframe and the frame are extracted by the noise source extracting step. And a multi-pulse codebook composed of a plurality of subframe-length code vectors composed of a plurality of pulses to extract and encode a noise source component of the subframe. And
In the noise source extraction step, if it is necessary to make a pulse at a non-integer sample position, a pulse is made at a plurality of integer sample positions sandwiching the non-integer sample position, so that noise source extraction can be performed with high accuracy. Can do
A speech coding method that can achieve higher sound quality can be obtained.

According to the invention of claim 6, according to claim 5,
In the invention of the above, when it is necessary to raise a pulse at the non-integer sample position α in the noise source extracting step, when the fractional part of α is β (0 <β <1), the largest integer sample position ( A pulse having an amplitude of (1-β) times the pulse amplitude is set at (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. Therefore, it is possible to obtain a speech coding method capable of extracting a noise source with high accuracy and a high sound quality with a simpler configuration.

According to the seventh aspect of the present invention, the digital audio signal is divided into frames in the frame division step, the frame is further divided into shorter subframes in the subframe division step, and the frame is extracted from the frame in the formant parameter extraction step. When the formant parameters are extracted and encoded, the pitch period of the subframe is extracted and encoded using the subframe and the formant parameter in the pitch period extraction step, and the formant parameter and the pitch of the subframe and the frame are extracted and encoded in the noise source extraction step. A noise source component of a sub-frame is extracted and encoded using a period and a multi-pulse codebook composed of a plurality of code vectors having a sub-frame length composed of a plurality of pulses. Then, in the pitch period extracting step, not only the integer pitch period but also a non-integer pitch period is extracted, and the noise source extracting step includes a pitch periodizing step of periodicizing the code vector corresponding to the pitch period. If a pulse needs to be raised at a non-integer sample position in the periodization process, a pulse is generated at a plurality of integer sample positions sandwiching the non-integer sample position. Can be performed with high precision, and a speech coding method that can achieve higher sound quality can be obtained.

The invention according to claim 8 is the same as the invention according to claim 7.
In the invention of the above, if it is necessary to raise a pulse at the non-integer sample position α in the pitch periodization step, and when the fractional part of α is β (0 <β <1), the largest integer sample position smaller than α ( A pulse having an amplitude of (1-β) times the pulse amplitude is set at (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. Therefore, it is possible to obtain a speech encoding method capable of performing a pitch period with respect to a non-integer pitch of a multi-pulse type sound source with high accuracy and a high sound quality with a simpler configuration.

According to the ninth aspect of the present invention, the digital audio signal is divided into frames in the frame division step, the frame is divided into shorter subframes in the subframe division step, and the frame is divided from the frame in the formant parameter extraction step. When the formant parameters are extracted and encoded, the pitch period of the subframe is extracted and encoded using the subframe and the formant parameter in the pitch period extraction step, and the formant parameter and the pitch of the subframe and the frame are extracted and encoded in the noise source extraction step. A noise source component of a sub-frame is extracted and encoded using a period and a multi-pulse codebook composed of a plurality of code vectors having a sub-frame length composed of a plurality of pulses. The pitch period extracting step extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extracting step includes a pitch periodizing step for periodicizing the code vector in accordance with the pitch period. If it is necessary to make a pulse at a non-integer sample position in the periodization step, make a pulse at a plurality of integer sample positions sandwiching the non-integer sample position, so that the pulse with respect to the non-integer pitch of the multi-pulse sound source can be obtained. The pitch cycle can be performed with high precision, and the voice coding algorithm that can improve the sound quality is recorded on a recording medium, so that the voice coding algorithm becomes machine-readable, and these operations can be performed by a computer. A recording medium that can be realized can be obtained.

According to a tenth aspect of the present invention, in the ninth aspect of the present invention, when it is necessary to set a pulse at a non-integer sample position α in the pitch period setting step, the fractional part of α is β (0 < When β <1), the pulse amplitude (1−1) is set at the maximum integer sample position (α−β) smaller than α.
β) A pulse having an amplitude that is twice as large is set, and a pulse having an amplitude that is β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) that is larger than α. Performing the pitch period for the non-integer pitch of the sound source with high precision and recording the voice encoding algorithm capable of improving the sound quality on a recording medium, the voice encoding algorithm becomes machine-readable,
A recording medium capable of realizing these operations by a computer can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of a speech encoding device according to the present embodiment.

FIG. 2 is an explanatory diagram of a case where a pulse pitch cycle is performed when P <α <M.

FIG. 3 is a diagram illustrating a case where a pulse is generated at a non-integer sample position when α is a non-integer and a plurality of pulses are generated by approximating the pulse to a triangular wave and interpolating the pulse to substitute for the pulse at the non-integer sample position FIG.

FIG. 4 is a diagram illustrating that a small number of pulses can be approximately substituted even when it is necessary to raise a pulse at a non-integer sample position by approximating a pulse to a triangular waveform and interpolating the pulse. .

FIG. 5 is a flowchart showing an example of an operation of encoding a digital audio input signal.

FIG. 6 is a flowchart illustrating another operation example of the encoding processing of the digital audio input signal.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Speech coding apparatus 12 Frame division part 14 Subframe division part 16 Formant parameter extraction part 18 Pitch period extraction part 20 Noise source extraction part 22 Target signal construction part 24 Pitch periodization part 26 Distance calculation part 28 Optimal vector determination part 20 Gain Quantization unit

Claims (10)

[Claims] 1. An audio encoding device for encoding a digital audio signal, comprising: a frame dividing unit that divides the digital audio signal into frames; a subframe dividing unit that divides the frame into shorter subframes; A formant parameter extracting unit that extracts and encodes a formant parameter from a frame; a pitch period extracting unit that extracts and encodes a pitch period of the subframe using the subframe and the formant parameter; A noise source for extracting and encoding a noise source component of the subframe using a multi-pulse codebook constituted by a plurality of code vectors having a subframe length constituted by a formant parameter, the pitch period, and a plurality of pulses. Extraction means, and the noise source In detecting means, when the need to make a pulse to a non-integer sample positions occurs, speech coding apparatus characterized by make a pulse integer sample positions of a plurality of locations across the non-integer sample positions. 2. When the noise source extracting means needs to generate a pulse at a non-integer sample position α, when the fractional part of α is β (0 <β <1), the largest integer smaller than α A pulse having an amplitude of (1-β) times the pulse amplitude is set at the sample position (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. The speech encoding device according to claim 1, wherein the speech encoding device is set up. 3. An audio encoding apparatus for encoding a digital audio signal, comprising: a frame dividing unit that divides the digital audio signal into frames; a subframe dividing unit that divides the frame into shorter subframes; A formant parameter extracting unit that extracts and encodes a formant parameter from a frame; a pitch period extracting unit that extracts and encodes a pitch period of the subframe using the subframe and the formant parameter; A noise source for extracting and encoding a noise source component of the subframe using a multi-pulse codebook constituted by a plurality of code vectors having a subframe length constituted by a formant parameter, the pitch period, and a plurality of pulses. Extraction means, and A pitch period extracting unit that extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extracting unit includes a pitch periodizing unit that periodicizes the code vector in accordance with the pitch period; A speech encoding apparatus characterized in that, when it is necessary for a pitch periodizing means to generate a pulse at a non-integer sample position, a pulse is generated at a plurality of integer sample positions sandwiching the non-integer sample position. 4. When the pitch period generation means needs to generate a pulse at a non-integer sample position α, when the fractional part of α is β (0 <β <1), the largest integer smaller than α A pulse having an amplitude of (1-β) times the pulse amplitude is set at the sample position (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. The speech encoding device according to claim 3, wherein the speech encoding device is set up. 5. A speech encoding method for encoding a digital speech signal, comprising: a frame division step of dividing the digital speech signal into frames; a subframe division step of dividing the frame into shorter subframes; A formant parameter extraction step of extracting and encoding a formant parameter from a frame; a pitch period extraction step of extracting and encoding a pitch period of the subframe using the subframe and the formant parameter; and A noise source for extracting and encoding a noise source component of the subframe using a multi-pulse codebook constituted by a plurality of code vectors having a subframe length constituted by a formant parameter, the pitch period, and a plurality of pulses. Extracting the noise source In out step, if necessary to make a pulse to a non-integer sample positions occurs, speech coding method, characterized in that to make a pulse integer sample positions of a plurality of locations across the non-integer sample positions. 6. When the fractional part of α is β (0 <β <1) when a pulse needs to be raised at a non-integer sample position α in the noise source extraction step, the largest integer smaller than α A pulse having an amplitude of (1-β) times the pulse amplitude is set at the sample position (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. The speech encoding method according to claim 5, wherein the speech encoding is performed. 7. A speech encoding method for encoding a digital audio signal, comprising: a frame dividing step of dividing the digital audio signal into frames; a subframe dividing step of dividing the frame into shorter subframes; A formant parameter extraction step of extracting and encoding a formant parameter from a frame; a pitch period extraction step of extracting and encoding a pitch period of the subframe using the subframe and the formant parameter; and Noise that extracts and encodes a noise source component of the subframe using a formant parameter, the pitch period, and a multi-pulse codebook composed of a plurality of code vectors having a subframe length composed of a plurality of pulses. A source extraction step; and The pitch period extracting step extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extracting step includes a pitch periodizing step of periodicizing the code vector in accordance with the pitch period, The voice coding method according to claim 1, wherein, in the pitch period setting step, when a pulse needs to be set at a non-integer sample position, a pulse is set at a plurality of integer sample positions sandwiching the non-integer sample position. 8. In the case where it is necessary to raise a pulse at a non-integer sample position α in the pitch period forming step, when the fractional part of α is β (0 <β <1), the largest integer smaller than α A pulse having an amplitude of (1-β) times the pulse amplitude is set at the sample position (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. The speech encoding method according to claim 7, wherein the speech encoding is performed. 9. A computer-readable recording medium recording an audio encoding algorithm for encoding a digital audio signal, comprising: a frame dividing step of dividing the digital audio signal into frames; A subframe dividing step; a formant parameter extracting step of extracting and encoding a formant parameter from the frame; and a pitch period of extracting and encoding a pitch period of the subframe using the subframe and the formant parameter. The extracting step, the sub-frame, the formant parameters, the pitch period and the sub-frame of the sub-frame using a multi-pulse codebook composed of a plurality of code vectors of the sub-frame length composed of a plurality of pulses A noise source extraction step of extracting and encoding a sound source component, further comprising: the pitch period extraction step extracts not only an integer pitch period but also a non-integer pitch period, and the noise source extraction step A pitch cycling step of cycling the code vector in accordance with the pitch cycle, wherein the pitch cycling step sandwiches the non-integer sample position when it is necessary to raise a pulse at a non-integer sample position. A computer-readable recording medium on which a speech encoding algorithm is characterized in that pulses are formed at a plurality of integer sample positions. 10. In the case where it is necessary to raise a pulse at a non-integer sample position α in the pitch periodization step, when the fractional part of α is β (0 <β <1), the largest integer smaller than α A pulse having an amplitude of (1-β) times the pulse amplitude is set at the sample position (α-β), and a pulse having an amplitude of β times the pulse amplitude is set at the smallest integer sample position (α-β + 1) larger than α. A computer-readable recording medium on which the voice encoding algorithm according to claim 9 is recorded.