JP2002099300A - Method and device for coding voice - Google Patents

Abstract

PROBLEM TO BE SOLVED: To code a voice pitch which is a voice coding parameter of analytico- synthetic type, vocal/silent information and amplitude of harmonics spectrum with a small number of bits. SOLUTION: The voice pitch is quantized by switching a frame of logarithmic pitch, which is quantized by uniform/differential selection, to a frame, which is quantized by an interpolating pitch candidate index between the frames, by an input switching part 103. The vocal/silent information is quantized by switching a frame, which is quantized by an index of the vocal/silent information selected from representative vocal/silent information 132, to a frame, which does not transmit the vocal/silent information, by a frame thinning part 133. The amplitude of harmonics spectrum is quantized by switching a frame, which is modeled by an autoregressive linear predictive model adjusted in spectrum, to a frame which is quantized by the index of interpolating candidate of modeling coefficient, by an input switching part 141. As a result, the voice coding parameter can be satisfactorily quantized with low bit rate.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoding method and apparatus for digitizing an audio signal and encoding an audio encoding parameter representing its characteristics at predetermined time intervals. Digital mobile phones or digital voice storage that transmits or stores voice coding parameters, restores voice coding parameters from the transmission destination or storage destination when necessary, synthesizes voice signals from the recovered voice coding parameters, and transmits voice. It is suitable for use in devices and the like.

[0002]

2. Description of the Related Art Digitized audio signals can be subjected to various digital signal processing such as data compression, error processing, and multiplexing. Widely adopted. To digitize analog audio signals,
Generally, sampling is performed at a sampling frequency that is twice or more the input voice frequency band, and quantization is required in a quantization step that cannot be discerned by ears. Therefore, a wider transmission frequency bandwidth is required compared to analog signals. Therefore, the audio signal once digitized is subjected to data compression by various coding methods and modulation methods according to the required sound quality.

As a method of obtaining a high compression ratio of voice data, an analysis-synthesis type voice coding method that positively uses characteristics of voice and a voice coding parameter obtained therefrom are efficiently quantized. A way is being considered. For example, MBE (Multi-
Band Excitation) or IMBE (Improved Mu)
lti-Band Excitation) is a type of analysis-synthesis-type speech coding system in which speech is transmitted at a predetermined time interval (20 msec).
, A frame is formed by dividing into segments, and for each frame, a voice pitch (or a voice fundamental frequency as a reciprocal thereof), a voice harmonics spectrum amplitude sequence obtained from the frequency spectrum of the voice of the frame, and a frequency spectrum of an appropriate frequency Voiced / unvoiced information (Voiced / Unvoiced information or V / UV information) for each frequency band divided into regions is used as a voice coding parameter. For each frame, the voice pitch is 8-bit uniform quantization, and V for each band. / UV information v
[k] (k is the number of a band) is a binary value represented by a binary number of 0/1, and is K-bit quantized (K: 12 at the maximum number of bands)
(Variable length of bits), and the speech harmonic amplitude sequence obtains a speech encoding speed of 4.15 kbps by two-dimensionally transforming the inter-frame prediction difference value and quantizing its DCT (discrete cosine transform) coefficient by 75-K bits.

FIG. 7 is a diagram showing a configuration of a general speech coded transmission apparatus. The sampled and quantized audio digital signal input from the audio input terminal 301 is divided into segments at predetermined time intervals by an audio encoding parameter extraction unit 302 to form frames, and audio encoding is performed for each frame. Extract parameters. The speech coding parameters to be extracted differ depending on the speech coding method.
In the BE system, it is a voice pitch, a voice harmonic spectrum amplitude sequence, and V / UV information of each frequency band. The parameter encoding unit 303 effectively encodes the extracted speech encoding parameters to reduce the code amount, and
To the transmission path 305 via The signal received by the receiving unit 306 is used to restore the speech coding parameters by the parameter decoding unit 307, and the speech synthesis unit 308 creates a synthesized speech by the reverse operation of the speech coding parameter extraction unit 302, and outputs the speech from the speech output terminal 309. Outputs audio digital signal.

FIG. 8 is a block diagram of the speech coding parameter extraction unit 302 in the case of the MBE system. The digital input audio signal is input from an input terminal 301 to a fundamental frequency estimating unit 401, where the fundamental frequency of the audio is estimated. The estimated value of the fundamental frequency is calculated as the reciprocal value of the time at which the autocorrelation function of the time delay is maximized. The frequency spectrum calculation unit 402 obtains a voice frequency spectrum by frequency-analyzing a finite-length voice signal cut out from a frame using a window function such as a Hamming window. The fundamental frequency correction unit 403 performs an AbS (Analysis) under a minimum error condition between the estimated speech fundamental frequency, the spectrum synthesized by the window function, and the speech frequency spectrum.
-by-Synthesis) method to simultaneously obtain the modified voice fundamental frequency ωo and the harmonics spectrum amplitude sequence. The voiced strength calculation unit 404 divides the frequency band into a plurality of frequency bands k (k = 1, 2,..., K) based on the modified voice fundamental frequency ωo, and synthesizes a synthesized spectrum for each frequency band. And the difference between the voice frequency spectrum and the threshold
Outputs V / UV information v [k]. Spectrum envelope calculator 40
5 is based on V / UV information v [k].
In the unvoiced band, the root-mean-square value of the frequency spectrum in each harmonics frequency band is calculated as the spectral envelope absolute value | A (ω) |
Output as

FIG. 9 is a block diagram of the parameter encoding unit 303 in the case of the IMBE system.
The fundamental voice frequency ωo input to the input terminal 501 is uniformly quantized to 8 bits in a predetermined quantization range and a predetermined quantization step by the basic frequency quantization unit 502, and the quantized value B0 is output to the output terminal 503. I do. The V / UV information v [k] input to the input terminal 504 is output to the V / UV information quantization unit 505.
For example, when the number of frequency bands K is 12, twelve 0s
Alternatively, it outputs to the output terminal 506 as a binary 12-bit value B1 represented by the information of 1. The spectrum envelope | A (ω) | input to the input terminal 507 is a discrete harmonics spectrum amplitude sequence | A (ωi) |, (i = 1,2,3, ..., N, N: number of harmonics ). First, after the logarithmic conversion unit 508 performs logarithmic conversion, a subtractor 509 calculates a difference from a predicted harmonics spectrum amplitude sequence 521 predicted from the harmonics spectrum amplitude sequence of the previous frame (this is referred to as a “prediction difference value sequence”). Then, it is passed to the block conversion unit 510. The block transform unit 510 sequentially classifies the prediction difference value sequence into six types according to the order of harmonics to form two-dimensional data, passes the resulting data to a DCT (discrete cosine transform) unit 511, calculates DCT coefficients, and passes the DCT coefficients to a quantization unit 512. , DC
The quantized data B2 of the prediction difference value sequence is output to the output terminal 513 by a combination of the uniform quantization method and the vector quantization method preliminarily selected according to the order of the T coefficient. Quantization restoration section 514, inverse DCT section 515, and block restoration section 516
Restores the quantized sequence of prediction difference values, and
Is added to the predicted harmonics spectrum amplitude sequence 521,
The quantized spectrum envelope value of the input spectrum envelope of the current frame is restored. The quantized spectrum envelope value is delayed by one frame in the frame delay unit 518, and the spectrum envelope prediction unit 519 uses the spectrum fundamental prediction unit 519 based on the newly input speech fundamental frequency ωo of the next frame and the fundamental frequency of the previous frame. A value is predicted, and the predicted value is led to the subtractor 509 to prepare for quantization of the spectral envelope of the next frame.

The principle of the MBE system is based on DW Griffin
and JS Lim, "Multi-band Excitation Vocoder", I
EEE Transactions on Acoustics, speech, and signal
processing, vol. 36, No. 8, August 1988, pp 1223-1235. Also, the encoding method of the encoder is USP-5491722 (Methods for
speech transmssion, Feb. 13, 1996). As described above, an analysis-synthesis-type speech coding scheme that extracts and encodes speech coding parameters based on a speech synthesis model has been proposed as a method of realizing low bit rate speech coding by digitizing speech. , Some of which are in practical use.

[0008]

The above-described analysis-synthesis type speech coding method is effective for low bit rate speech coding. However, this method converts speech based on a certain speech analysis / synthesis model. Since the speech is analyzed and synthesized only with the speech synthesis parameter, the sound quality tends to be a synthesized sound depending on the configuration of the encoding method. As a method of improving this point, a method of setting a short speech frame update period to reduce a change in speech parameters in the speech frame and improving the sound quality while being of the analysis-synthesis type can be considered. Hokuto et al., "Study on speech coding for commercial mobile communications", IEICE 2000 National Convention, D14-2, p171 , Mar.2000.

[0009] However, shortening of the update period of the audio frame goes against high compression ratio (low bit rate) of audio. For example, in the above report of Fumoto et al., A speech frame is divided into two and a speech coding parameter is taken out as a subframe structure. If the speech coding parameter is encoded by a conventional method, a double bit rate is required. Becomes Therefore, a parameter encoding method in which the number of encoded bits of the parameter does not become extremely large even when the audio segment length is set short is desired. As described above, in order to improve the voice quality of the analysis-synthesis-type speech coding scheme aiming at a low bit rate, there is a problem of an efficient coding method of the analysis-synthesis-type speech coding parameter. There is a problem in a measure for preventing an increase in the encoding bit rate when is shortened.

Accordingly, an object of the present invention is to provide a speech encoding method and apparatus capable of greatly reducing the encoding bit rate in an analysis-synthesis speech encoding method and apparatus. Also, in the analysis-synthesis type speech coding method and apparatus, the analysis-synthesis-type speech coding method improves the quality of the coded speech by increasing the frame update period of the speech coding and prevents the number of coded bits from increasing. It is an object of the present invention to provide a speech encoding method and apparatus.

[0011]

In order to solve the above-mentioned problems, a voice coding method according to the present invention obtains voice coding parameters from a voice signal digitized and divided into frames of a predetermined time length, and performs coding. Speech encoding method,
The voice pitch as the voice coding parameter, a plurality of frames calculated by using the quantized voice pitch of a frame to be quantized by any one of the quantization methods by selecting the difference quantization method and the uniform quantization method, and the preceding and succeeding frames. Selecting one of the interpolated speech pitch candidates of the above, and encoding by combining with a frame to be quantized by the index number,
Encoding the unvoiced information by the index number of the nearest one of a limited number of representative voiced / unvoiced information, and a harmonics spectrum amplitude sequence as the speech encoding parameter is calculated by a linear prediction model. A frame to be quantized by a linear prediction coefficient (or a line spectrum pair derived therefrom) and a gain, and a plurality of interpolated linear prediction coefficients (or line spectra) calculated using the linear prediction coefficients (or line spectrum pairs) of the preceding and succeeding frames. Pair) selecting one of a plurality of combinations of the interpolation gain candidates calculated using the gains of the candidate and the preceding and succeeding frames, and performing encoding by a combination of frames to be quantized by the index number.

In the quantization of the voice pitch, the difference quantization method and the uniform quantization method are selected. When the difference quantization error is equal to or less than a certain threshold value, the difference quantization method is selected. In the process of selecting a quantization method with a small quantization error, the selection of the interpolation voice pitch candidate is performed by selecting the interpolation voice pitch closest to the voice pitch of the current frame among a plurality of interpolation voice pitch candidates calculated from the voice pitches of the previous and next frames. The process is to select a voice pitch candidate. Furthermore, the quantization of the voice pitch is performed using a logarithmic pitch obtained by logarithmically converting the voice pitch. Furthermore, the difference quantization of the voice pitch is performed by setting the quantization step to be larger as the difference value increases.

Further, the limited number of representative voiced / unvoiced information is deleted from a large number of previously obtained voiced / unvoiced information in order of frequency of occurrence, and the deleted voiced / unvoiced information is deleted. The generation frequency is assumed to be created by reducing the number of occurrences to a desired limited number of voiced / unvoiced information by distributing the occurrence frequency to adjacent voiced / unvoiced information according to the distance and size to the information, and integrating them. I have. Furthermore, the voiced / unvoiced information distance represents voiced / unvoiced for each voice spectrum band as 0 or 1, and the value of 0 or 1 of each voice spectrum band is converted into a binary bit in the frequency of the voice spectrum band. It is supposed to be applied and represented with the difference between the binary values. Furthermore, the voiced / unvoiced information is constituted by a combination of a frame for transmitting voiced / unvoiced information and a frame not to be omitted and transmitted. Furthermore, in a frame from which the voiced / unvoiced information is omitted, decoding is performed using voiced / unvoiced information of a frame having a large energy or an average amplitude value of the frames before and after the frame. Furthermore, when the harmonics spectrum amplitude sequence is modeled by a linear prediction model, a linear prediction model of the harmonics spectrum amplitude is performed after correcting the harmonics amplitude value of the 0th order (DC component). . Still further, the plurality of interpolation gain candidates include gains obtained by equally or unequally dividing the gains of the preceding and succeeding frames, and gains of the larger and smaller gains of the preceding and following frames. .

Still further, the speech encoding apparatus of the present invention comprises:
An audio encoding device that acquires and encodes an audio encoding parameter from an audio signal that has been digitized and divided into frames of a predetermined time length, wherein an audio pitch as the audio encoding parameter is equalized with a differential quantization method. A frame to be quantized by any of the quantization methods depending on the selection of the quantization method;
Means for selecting any of a plurality of interpolated speech pitch candidates calculated using the quantized speech pitches of the preceding and succeeding frames, and encoding by a combination of frames to be quantized by the index number; Means for encoding voiced / unvoiced information as an index number of a limited number of representative voiced / unvoiced information at the closest distance from the limited number of representative voiced / unvoiced information; A frame to be quantized by a linear prediction coefficient (or a line spectrum pair derived therefrom) and a gain by a prediction model, and a plurality of interpolated linear prediction coefficients (or a line spectrum pair calculated using the linear prediction coefficients (or a line spectrum pair) of the preceding and following frames) Or line spectrum pair) using the gains of the candidate and the previous and next frames. Select one of the combinations of the plurality of interpolated gain candidate, by its index number is intended to include means for encoding by the combination of a frame to be quantized.

As described above, in the present invention, the coding of the voice pitch is performed with respect to the logarithmically converted pitch.
Switching between the difference quantization method and the uniform quantization method, and selecting the one with the smaller error from the input voice pitch to quantize;
By selecting the number of the nearest interpolation point candidate from a plurality of interpolation points of the inter-frame audio pitch and quantizing with the selected number, the number of quantization bits of the audio pitch can be reduced by switching and using the frame repetition. is decreasing. For encoding of voiced / unvoiced information (V / UV information),
The V / UV information and its occurrence frequency are obtained in advance for many voice frames, a fixed number of representative V / UV information is selected from among them, and the V / UV of each frame is selected from the representative V / UV information. Means is used for encoding with the number (index) of the representative V / UV information most similar to the information. Furthermore, a frame that does not transmit V / UV information is appropriately inserted, and decoding of a frame for which V / UV information is not transmitted is performed for decoding a frame having higher sound energy among the preceding and following frames. V / UV information is used. Furthermore, regarding encoding of the harmonics spectrum amplitude sequence, the spectrum amplitude sequence is modeled by a higher-order all-pole model, that is, an autoregressive linear prediction model (AR model), and its linear prediction coefficient (L
(PC coefficient) and gain, or LSP (line spectrum pair) obtained by transforming the LPC coefficient and a means for quantizing the gain. Also, the LPC coefficient of the frame or L
For the quantization of SP and gain, a means for selecting the number of the nearest interpolation point candidate from a plurality of interpolation points between the quantized LPC coefficient or LSP and gain frame and quantizing with the selected number is also used. Thereby, the number of quantization bits of the harmonics spectrum amplitude sequence is reduced.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of a speech encoding method and apparatus according to the present invention will be described by taking as an example a case where the speech encoding method is applied to the MBE or IMBE speech encoding method which is the analysis-synthesis speech encoding method. Note that this speech encoding device corresponds to the parameter encoding unit 303 shown in FIG. 7, and the speech encoding parameters extracted by the speech encoding parameter extraction unit 302, that is, the speech pitch (or the reciprocal thereof) , The voice harmonic frequency spectrum amplitude sequence and voiced / unvoiced information (V / UV information) of each frequency band are efficiently encoded.

FIG. 1 is a block diagram showing an example of the configuration of a speech coding apparatus to which the speech coding method of the present invention is applied.
For example, the voice pitch (or voice fundamental frequency ωo) obtained by the voice coding parameter extraction unit shown in FIG. 8 is input to the input terminal 101, and the voice pitch is logarithmically converted by the logarithmic conversion unit 102, and the logarithmic voice pitch P [n] (n is the frame number). Logarithmic voice pitch is described in the literature (Thomas Eriksson
and Hong-Goo Kang, "Pitch Quantization in low Bit-
Rate Speech Coding ", ICASSP '99, pp489-492, 1999)
It is known that the detection limit value of a person with respect to the change amount of the logarithmic pitch is not significantly affected by the value of the logarithmic pitch as described in US Pat. Therefore, since the quantization step width can be made uniform, the conversion is convenient.

The logarithmic pitch P [n] is alternately switched for each frame (or for each subframe when subframes are formed) by the input switching unit 103, and the two output terminals 1
04 or 116 is output. When output to 104, it is guided to uniform quantization section 105 and subtraction section 112. The uniform quantization unit 105 performs uniform quantization at a fixed quantization step, and the quantized logarithmic pitch P1 ′ [n] is input to the pitch comparison unit 108. On the other hand, the subtraction unit 112 obtains the difference logarithmic pitch from the input logarithmic pitch and the quantization logarithmic pitch of the previous frame received from the delay unit 111 and inputs the difference logarithmic pitch to the difference quantization unit 113. The delay unit 111
This is for passing the quantized logarithmic pitch of the immediately preceding frame to the current frame. The difference quantization unit 113 performs the difference quantization in a fixed difference quantization step or a non-uniform quantization step set such that the zero input is symmetric and the difference quantization step is expanded as the difference input amplitude increases,
The adder 114 adds the quantized logarithmic pitch of the previous frame as a reference, and sets the logarithmic pitch P2 ′ [n] by differential quantization to 10
7 is output.

In the pitch comparing section 108, P1 '[n] and P2' [n]
And selects the quantized logarithmic pitch having a smaller error from the logarithmic pitch P [n] before quantization, and outputs the quantized logarithmic pitch P ′ [n] of this frame to the output terminal 109. The output terminal 110 outputs, as a pitch code, the index output by the quantizer selected by the pitch comparison unit 108 from the uniform quantization index N1 and the differential quantization index N2. By arranging the indexes of N1 and N2 so that the numbers do not overlap, it is possible to determine which quantization method has been selected from the output index numbers.

The other output 116 of the input switching unit 103
The plurality of interpolation pitch candidates created by the interpolation pitch candidate creation unit 117 using the quantized logarithmic pitches of the current frame and the previous frame obtained from the input / output end of the delay unit 111 are The interpolation point index (selection number) N3 which is compared by the interpolation point comparison unit 119 and gives the pitch closest to the logarithmic pitch of the output 116 is 1 as the pitch interpolation code.
20.

FIG. 2 shows an interpolation pitch candidate creating section 11 shown in FIG.
FIG. 7 is a diagram for explaining the operation of No. 7; The example shown in FIG. 2 is an example in which the number of interpolation point candidates is 4 and quantization is performed with 2 bits according to the selection number. The quantized logarithmic pitch of the front frame of the current frame is P '[n + 1], and the quantized logarithmic pitch of the rear frame is P' [n-1]. Is indicated by a cross. Of the four interpolation pitch candidates, the interpolation quantization pitch P ′ closest to the input logarithmic pitch P [n]
[n] is selected, and 2 is selected as an index for giving the interpolation quantization pitch in the example of FIG. P [n] is P '[n
+1] and P '[n-1], for example,
When the frame is divided into two subframes, P [n] is the first subframe of the current frame, P ′ [n + 1] is the quantization pitch of the second subframe of the current frame, and P ′ [ n-
1] corresponds to the quantization pitch of the second sub-frame of the previous frame.

In the arrangement of the interpolation pitch candidates shown in FIG.
Although [n + 1] and P '[n-1] are not included in the interpolation pitch candidate, set the interpolation pitch candidate including P' [n + 1] and P '[n-1] at both ends. You can also. In that case, there are two interpolation pitch candidates other than P '[n + 1] and P' [n-1]. When the positions of the interpolation pitch candidates are set except for both ends as in the example of FIG. 2, two points excluding both ends can be selected even with one bit. Therefore, the interpolation point arrangement in FIG. Bit or 2
It can be said that it is effective when the number of bits is small.

Returning to FIG. 1, voiced / unvoiced information (V / UV
) Is input from the input terminal 131, and the frame thinning unit 133 thins the V / UV information. For example, 2
V / UV information is output only once for a frame,
It is input to the unvoiced comparing section 134. On the receiving side (decoding side), in a frame in which the voiced / unvoiced determination information is omitted, voiced / unvoiced information of a frame having a large frame energy or amplitude average value is used among frames before and after the frame. Decrypt.

The representative voiced / unvoiced information codebook 132 is one in which a limited number of frequently occurring information is selected from the V / UV information previously obtained from many voice frames and stored by the method according to the present invention described later. V / UV currently input
Voiced the representative V / UV information value b1 'closest to the information value b1 /
The unvoiced comparison unit 134 selects and outputs the index of the representative V / UV information as a voiced / unvoiced code 135. The V / UV information value b1 (or b1 ') is a V / UV information value v [k], k = 1, 2,... For each frequency band obtained by dividing the audio frequency spectrum at intervals of, for example, three times the audio fundamental frequency. .., K (v [k] is 0 or 1) is represented by a binary value obtained by allocating each bit of a binary number.

(Equation 1) The distance between the V / UV information values is represented by the absolute value of the difference between b1 and b1 '. The representative V / UV information is set independently for each band number determined by the audio fundamental frequency.

The representative V / UV information can be selected and created by the following method. That is, a large number of V / UV information values are obtained in advance from a large number of audio frames and are classified for each band number K. A set of V / UV information values classified for each band number {b1 _i }
, The frequency of occurrence of each b1 _i is tabulated. For example, 5 bands
For the case, the value of b1 takes an integer value of 0 to 31,
The occurrence frequency is totaled for each b1 value. When quantizing V / UV information with, for example, 2 bits from among them, it is necessary to select four types of representative V / UV information from among them. For this selection, a method of selecting four items in order from the highest occurrence frequency can be considered.
The UV information value may be selected, and a case where it is not appropriate as the representative V / UV information value occurs. In particular, when the number of bands K is large, there is a high possibility that the representative V / UV information value having a high frequency of occurrence exists adjacently.

Therefore, in the present invention, the occurrence frequency of the V / UV information having the lowest occurrence frequency is sequentially erased while distributing it to the occurrence frequency of the adjacent V / UV information, and finally, the representative V / UV information of the target number is finally erased. UV
We took the method of reducing until the number of information. Figure 3 shows V / UV
When sequentially deleting the number of information (V / UV pattern) candidates,
FIG. 9 is a diagram for explaining a method of erasing a V / UV pattern that occurs least frequently. Here, assuming that the occurrence frequency of the least frequently occurring V / UV pattern is q [n], q [n] is divided into q1 and q2, and the occurrence frequencies q [nl ₁ ] and q of the adjacent V / UV patterns, respectively.
[n + l ₂ ]. The distribution amount of q1 and q2 depends on the occurrence frequency q [nl ₁ ] and q [n + l ₂ ] of the adjacent V / UV pattern,
It is determined by the following formula according to the proximity of the distances l ₁ and l ₂ to the V / UV pattern.

(Equation 2)

Returning to FIG. 1, coding of the spectral envelope will be described. As the spectrum envelope, a discrete spectrum is input to the input terminal 140 as a harmonics spectrum amplitude sequence A [l], and the input switching unit 141 switches the path for each frame, and one is input to the spectrum correction unit 142. The operation of spectrum correcting section 142 will be described later. The spectrum interpolation unit 143 interpolates and generates the harmonics spectrum amplitude sequence input as the discrete spectrum amplitude sequence between the discrete spectra, and the linear prediction modeling unit 144 generates many spectrum amplitude sequences effective for modeling. In this spectrum interpolation, the harmonics spectrum amplitude is transformed into a logarithm to perform linear interpolation, and as a result, nonlinear interpolation is performed. Of course, other non-linear interpolation may be used, and an interpolation method that is well suited to the linear prediction model is desired.

The spectrum amplitude sequence whose number of samples has been increased by interpolation is modeled by an autoregressive linear prediction model in a linear prediction modeling unit 144, and gain g and a high-order (for example, up to 10th) linear prediction coefficient (LPC) are calculated. Coefficient a [j], j = 1,
2,3, ..., J, where J is the predicted order). The gain g is converted into a logarithmic gain by the gain quantization unit 145 and is uniformly quantized.
The index is the spectral gain code output terminal 155
Is output to On the other hand, the LPC coefficient a [j] is converted by the modeling coefficient conversion unit 147 into a line spectrum pair (LSP) F [j]. The LSP has a value of 0 to π, a range of change is fixed, there is little deterioration in audibility due to linear interpolation, and the LSP is widely and generally used as a coefficient for modeling a spectral envelope. The LPC coefficient converted into the LSP is vector-quantized by the LSP quantizer 148 using the LSP codebook 149, and the index of the codebook is output to the spectrum envelope code output terminal 156 as a spectrum envelope code.

Here, the function and the purpose of the spectrum correcting section 142 will be described. When modeling a harmonics spectral amplitude sequence with a linear prediction model, it is better to have many spectral points, but if there is a spectrum that does not fit the model, distortion will be given to the modeling and the spectral error after modeling will increase. I do. Generally, the 0th-order harmonics spectrum amplitude (DC component) of voice is lower than other harmonics spectrum amplitudes, and a modeling error is likely to occur. Also, since the 0th-order harmonics spectrum amplitude is an unnecessary component at the time of audio decoding (the audio signal hardly contains a DC component), it can be said that the amplitude may be adjusted and changed to a level that is easy to model. For the above reasons, the spectrum correction unit 142 sets the 0
Modify the next harmonics spectrum amplitude. Specifically, the value calculated by the following equation is replaced with the zero-order harmonics amplitude.

(Equation 3) Here, H ′ ₀ is the corrected zero-order harmonics spectrum amplitude, and H ₁ , H _{2, and} H ₃ are the first-order, second-order, and third-order harmonics amplitudes, respectively. It should be noted that the zero-order harmonics amplitude may be replaced using an equation other than this equation.

On the other hand, the other input harmonics spectrum amplitude sequence A [l] switched for each frame by the input switching unit 141 is compared with a spectrum interpolation point comparison unit 154.
Is input to In the spectrum interpolation point comparison unit 154, an interpolation gain candidate obtained by the interpolation calculation of the quantization gain of the preceding and succeeding frames from the input / output terminal of the gain delay unit 146 in the gain interpolation calculation unit 150, and the LSP interpolation calculation unit 152 The harmonics spectrum amplitude A ′ _i [l] (l is restored by the spectrum restoration unit 153 from the combination of the candidate interpolation LSP coefficients obtained by the interpolation calculation of the LSP coefficients of the frames before and after from the input / output end of the delay unit 151. Harmonic number, i is an interpolation candidate number or index), and the inputted harmonics spectrum amplitude sequence A
An index i that gives the interpolation gain and the interpolation LSP coefficient with the least spectral error compared to [1] is selected by the following equation, and is output to the spectrum interpolation code output terminal 157 as a spectrum interpolation code.

(Equation 4) Here, the argmin_i (x) function (function in which i is written below argmin in the above equation) is a function that evaluates i as a parameter and returns i that minimizes x.

As for the interpolation gain candidates, inconvenience may occur if only interpolation candidates between the quantization gains of the preceding and succeeding frames are used. For example, if there is a sharp minimum in the gain, the coded speech cannot be reduced to a sufficiently small level, generating unpleasant noise. In order to avoid this, the problem is reduced by adding to the interpolation gain candidate a candidate having a quantization gain of the preceding and succeeding frames that is equal to or greater than the minimum value or a maximum value. Specifically, the maximum value (the larger value of the quantization gain of the preceding and succeeding frames) +5 dB, +10 dB, and the minimum value (the smaller value of the quantization gain of the preceding and succeeding frames) -5 dB, -1
0 dB and the like are added to the interpolation gain candidates obtained by the interpolation calculation of the quantization gains of the preceding and succeeding frames.

Next, the flow of the processing of the encoding unit shown in FIG. 1 according to the present invention will be described. FIG. 4 is a flowchart of encoding of the fundamental voice frequency ωo. The process is started from 701 in FIG. Speech fundamental frequency ω quantized in 702
o, set the frame number m. Next, the logarithmic pitch P is calculated in 703, and even or odd of m is determined in 704. If the number is even, uniform quantization and differential quantization are performed in 705 and 706, respectively, and the uniform quantization pitch P_u and the uniform quantization index are calculated. Index_u, differential quantization pitch P_d, and differential quantization index Index_d are calculated. Next, the difference quantization error (| P_d-P |) is determined in 708, and if it is smaller than a certain threshold Th, the pitch P [m] of the m-th frame (P [2n in the sense of an even frame) is determined in 710. ] Is represented by P_d, and its index is Index [m] (this is Index in the sense of an even frame)
[Represented by [2n]) is Index_d. On the other hand, at 708, | P_d-P
If it is determined that | is equal to or greater than Th, the uniform quantization error (| P_u−P |) is compared with the differential quantization error (| P_d−P |) in 707, and if the uniform quantization error is small, At 709, the pitch P [2n] of the m-th frame is P_u, and its index Index [2]
n] is set to Index_u, and conversely, the pitch P [2n] is set to P_ at 710.
Let d and Index [2n] be Index_d. In addition, the pitch P [2n] of the m-th frame is delayed by 2 frames at 711, and P [2n−
2], and is used as a reference logarithmic pitch for the difference quantization 706 when the next even-numbered frame is differentially quantized.

If m is an odd number in 704, pitch interpolation candidates are selected in 713. A pitch interpolation candidate is a set {Pinpol [2n-1] _i } (i = 0) of interpolation pitch candidates obtained by equally dividing the interval between the quantized logarithmic pitches of the preceding and succeeding frames into a plurality of pieces by the same method as shown in FIG. , 1, 2, 3,..., N−1, N are calculated at 712 as interpolation candidate points). The selection of the pitch interpolation candidate is performed using a logarithmic pitch P [2n-1] of m = 2n-1 frames.
Is selected, and its index number Index [2n-1] is set at 714 as the quantized pitch code of the odd-numbered frame. The pitch code in the case of the even frame is the index Index_d or Index_u of the quantization method selected in 709 or 710.
Index [2n] is similarly set at 714. The pitch codes for the two frames of the even and odd frames set in this way are output to 715. By using the encoding method of the fundamental voice frequency according to the present invention shown in FIG. 4, for example,
By using 4 bits for the pitch code and 1 bit for the pitch interpolation code, it is possible to satisfactorily encode the speech fundamental frequency for two frames.

FIG. 5 is a flowchart for encoding V / UV information.
Processing starts at 801. At 802, the voice fundamental frequency is set to ωo, the V / UV information value of the frame is set to v, and the frame number is set to m. In step 803, the number of bands K for ωo is obtained. K = (int) ((π−ωo / 2) / (ωo × B)) where B is the number of harmonics included in each band, and is determined in advance before encoding. used. At step 804, it is determined whether or not m is even. If m is even, at step 805, the group ｛VC for the number of bands K from the representative V / UV information value data 806 for each number of bands selected in advance.
In B [K] [i] UV, select the index i of the representative V / UV information value having a value closest to the most input V / UV value v. IndexV / UV [2n]
At 807, a voiced / unvoiced code is set. If the processing frame is an odd number, the voiced / unvoiced code is thinned out and not output. If the V / UV information encoding method according to the present invention shown in FIG. 5 is used, for example, two frames of V / UV information can be satisfactorily encoded using two bits for voiced / unvoiced code.

FIG. 6 is a flowchart of encoding a harmonics spectrum amplitude sequence. The processing is started from 901 and 90
In step 2, the harmonics spectrum amplitude is set to A [l]. The frame number is set to m. In step 903, whether the m is even or odd is determined. In the case of an even frame, in step 904, the 0th-order harmonics spectrum amplitude A [0] is corrected by multiplying the second-order harmonics spectrum amplitude A [2] by α. The method for determining the correction coefficient α is determined by the method described above. The harmonics spectrum amplitude sequence corrected in 905 is linearly interpolated by a logarithmic value to increase the number of spectrum sequences. At 906, the interpolated and increased spectrum sequence is modeled and represented as a value on a modeling curve in a linear prediction model. The gain g, which is a modeling coefficient of the linear prediction model, is uniformly quantized at 908 as a logarithmic gain to obtain a quantization gain g ′ and its index. The quantization step and quantization range are predetermined at 907. LP, the other modeling coefficient
The C coefficient a [j] is converted to a line spectrum pair (LSP) and converted to F
In step 909, the vector is quantized using the vector quantization table 910 to obtain a quantized vector F ′ [j] and its index.

The quantization gain g 'from 908 is delayed by two frames at 912, and a plurality of (M1) interpolation gain candidates between its input and output are subjected to logarithmic gain linear interpolation at 911.
It is led to 5. Also, the quantized LSP (F ′) from 909
[j]) is delayed by two frames at 913, a plurality of (M2) interpolated LSP candidates between the input and output are subjected to an LSP linear interpolation candidate operation at 914, and an LSP interpolation candidate sequence is led to 915. At 915, the interpolation harmonics spectrum A ′ [l] is restored from the combination (maximum M1 × M2) between the input LSP interpolation candidate and the gain interpolation candidate. At 916, a set of the restored interpolated harmonics spectrum trains and 903
Are sequentially compared, and the combination number that gives the interpolated harmonics spectrum sequence with the least spectral error is selected as the spectrum interpolation number. The spectrum gain code shown in FIG. 1 is obtained from 908, and the spectrum envelope code is obtained from 909, and the process ends at 917. With the encoding method of the harmonics spectrum amplitude sequence according to the present invention shown in FIG. Since the encoding can be performed well with the code and the spectrum interpolation code, the voice quality can be improved and the number of encoded bits can be reduced.

In the above description, the input switching units 103 and 141 shown in FIG. 1 are switched for each frame for easy understanding. However, the present invention is not limited to the switching for each frame. The operation of changing the switching cycle as needed can be easily realized. This causes slight changes in the flow charts of FIGS. 4, 5, and 6, but it is possible for related engineers and researchers to easily change the corresponding portions. Further, in the above-described embodiment, a frame that does not transmit voiced / unvoiced determination information is provided in order to further reduce the number of coded bits, but voiced / unvoiced information may be transmitted for all frames. Good. Further, in the above, the harmonics spectrum amplitude sequence is represented by a gain and a line spectrum pair (L
Although the quantization is performed by SP), the quantization may be performed by the gain and the linear prediction coefficient (LPC coefficient).

[0038]

As described above, according to the speech encoding method and apparatus of the present invention, the speech pitch (or speech fundamental frequency) and the V /
In an analysis-synthesis-type speech coding method represented by a speech coding parameter consisting of UV information and a harmonics spectrum amplitude sequence, in a speech pitch, a frame to be subjected to differential quantization or uniform quantization as a logarithmic pitch, and an inter-frame interpolation index. By performing encoding by switching frames to be quantized, the number of encoded bits can be significantly reduced. In addition, V / UV information is represented for each number of bands determined by the range of the sound fundamental frequency.
By encoding with V / UV value index numbers, the number of V / UV encoded bits can be reduced rationally. Furthermore,
The V / UV information is thinned out by frames, and the V / UV information of a frame that does not transmit V / UV information is further reduced in the number of V / UV encoded bits by a method of inferring from previous and subsequent frames described in the present invention. Can be done. Further, after correcting the 0th-order harmonics amplitude value of the voice harmonics spectrum amplitude sequence by the method described in the present invention, an autoregressive linear prediction model is formed, and the model coefficient is quantized and transmitted. Without interpolating, a plurality of interpolation model coefficients are obtained by inter-frame interpolation from model coefficients that have already been quantized, and coding is performed by switching frames to be quantized with an index number of an interpolation candidate that gives the best interpolation model coefficient. The number of coded bits of the harmonics spectrum amplitude sequence can be greatly reduced while improving the voice quality. As described above, according to the present invention, it is possible to provide a method and an apparatus for greatly reducing an encoding bit rate in an analysis and synthesis type speech encoding method and apparatus. Also, in the analysis-synthesis type speech coding method and apparatus, the analysis-synthesis-type speech coding method improves the quality of the coded speech by increasing the frame update period of the speech coding and prevents the number of coded bits from increasing. Can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an embodiment of a speech encoding apparatus to which a speech encoding method according to the present invention is applied.

FIG. 2 is a diagram for explaining an interpolation pitch candidate creation process in an interpolation pitch candidate creation unit of the present invention.

FIG. 3 is a diagram for explaining a method of deleting voiced / unvoiced information having a low frequency of occurrence in the present invention.

FIG. 4 is a flowchart of encoding a speech pitch in the speech encoding method of the present invention.

FIG. 5 is a flowchart of encoding voiced / unvoiced information in the speech encoding method of the present invention.

FIG. 6 is a flowchart of encoding a harmonics spectrum amplitude sequence in the speech encoding method of the present invention.

FIG. 7 is a configuration diagram of a speech coded transmission device.

FIG. 8 is a block diagram of a speech coding parameter extraction unit.

FIG. 9 is a block diagram of a parameter encoding unit in a conventional speech encoding scheme.

[Explanation of symbols]

102 logarithmic conversion unit, 103 input switching unit, 105 uniform quantization unit, 108 pitch comparison unit, 111 delay unit, 1
12 subtraction section, 113 difference quantization section, 114 addition section, 117 interpolation pitch candidate creation section, 119 interpolation point comparison section, 132 representative voiced / unvoiced information codebook, 133
Frame thinning section, 134 Voiced / unvoiced comparison section, 14
1 input switching unit, 142 spectrum correction unit, 143
Spectrum interpolating unit, 144 linear prediction modeling unit, 14
5 Gain quantization section, 146 gain delay section, 147 modeling coefficient conversion section, 148 LSP quantization section, 149 L
SP codebook, 150 gain interpolation calculator, 151
LSP delay section, 152 LSP interpolation calculation section, 153 spectrum restoration section, 154 spectrum interpolation point comparison section

Continuing from the front page (72) Inventor Seiji Sasaki 3-2 Hikarinooka, Yokosuka City, Kanagawa Prefecture F.R.P. BC14 BC16 BC26 BD01 5K041 AA00 CC01 DD01 EE22 EE24 EE31 HH23 HH43

Claims (11)

[Claims] 1. A speech encoding method for acquiring and encoding a speech encoding parameter from a speech signal digitized and divided into frames of a predetermined time length, wherein a speech pitch as the speech encoding parameter is obtained by subtracting Select a frame to be quantized by one of the quantization methods by selecting the quantization method or uniform quantization method, and select one of a plurality of interpolation voice pitch candidates calculated using the quantized voice pitch of the previous and next frames And coding by a combination with a frame to be quantized by the index number, wherein voiced / unvoiced information as the voice coding parameter is
Encoding the closest number of representative voiced / unvoiced information from the limited number of representative voiced / unvoiced information using an index number; and converting the harmonics spectrum amplitude sequence as the voice coding parameter into a linear prediction coefficient ( Or a frame to be quantized by a gain or a line spectrum pair derived therefrom, and a plurality of interpolation linear prediction coefficient (or line spectrum pair) candidates calculated using the linear prediction coefficients (or line spectrum pair) of the preceding and succeeding frames. Selecting one of a plurality of combinations of interpolation gain candidates calculated using the gains of the preceding and succeeding frames, and encoding the selected combination by a combination of frames to be quantized by the index number. Method. 2. The method according to claim 1, wherein the difference quantization method and the uniform quantization method in the quantization of the voice pitch are selected when the difference quantization error is equal to or smaller than a certain threshold value, and when the difference quantization error is equal to or larger than the threshold value, In the process of selecting a quantization method with a small quantization error, the selection of the interpolation voice pitch candidate is performed by selecting the interpolation voice pitch closest to the voice pitch of the current frame among a plurality of interpolation voice pitch candidates calculated from the voice pitches of the previous and next frames. 2. The speech encoding method according to claim 1, wherein said speech encoding method is a process of selecting a speech pitch candidate. 3. The speech encoding method according to claim 1, wherein the quantization of the speech pitch is performed using a logarithmic pitch obtained by logarithmically converting the speech pitch. 4. The speech encoding method according to claim 1, wherein the difference quantization of the speech pitch is performed by setting a quantization step larger as the difference value increases. 5. The limited number of representative voiced / unvoiced information is deleted from a large number of previously acquired voiced / unvoiced information in order of frequency of occurrence, and the generation of the removed voiced / unvoiced information is performed. By allocating and integrating the frequency to adjacent voiced / unvoiced information according to the distance and size to the information, it is created that the frequency is reduced to a desired limited number of voiced / unvoiced information. The speech encoding method according to claim 1, wherein: 6. The distance of the voiced / unvoiced information represents voiced / unvoiced for each voice spectrum band as 0 or 1, and a value of 0 or 1 of each voice spectrum band is represented by a binary number in the order of the frequency of the voice spectrum band. 2. A speech encoding method according to claim 1, wherein the method is applied to bits and represented with a difference between the binary values. 7. The speech coding method according to claim 1, wherein the voiced / unvoiced information includes a combination of a frame for transmitting voiced / unvoiced information and a frame not to be omitted and transmitted. 8. A frame in which the voiced / unvoiced information is omitted, wherein decoding is performed using voiced / unvoiced information possessed by a frame having a large energy or a large amplitude average value among frames before and after the frame. Item 7. A speech encoding method according to Item 7. 9. When the harmonics spectrum amplitude sequence is modeled by a linear prediction model, a zero-order (DC component)
2. The speech encoding method according to claim 1, wherein after correcting the harmonics spectrum amplitude value, linear prediction modeling of the harmonics spectrum amplitude is performed. 10. The plurality of interpolation gain candidates include gains obtained by equally or unequally dividing the gains of the preceding and succeeding frames, gains of the larger and smaller gains of the preceding and succeeding frames. The speech encoding method according to claim 1, further comprising: 11. A speech encoding apparatus for acquiring and encoding a speech encoding parameter from a speech signal digitized and divided into frames of a predetermined time length, wherein a speech pitch as the speech encoding parameter is obtained by subtracting Select a frame to be quantized by one of the quantization methods by selecting the quantization method or uniform quantization method, and select one of a plurality of interpolation voice pitch candidates calculated using the quantized voice pitch of the previous and next frames Means for encoding by a combination of frames to be quantized by the index number; and voiced / unvoiced information as the speech encoding parameter,
Means for encoding with a index number of the nearest one of a limited number of representative voiced / unvoiced information; and a linear prediction coefficient (or A frame to be quantized by the line spectrum pair derived therefrom and the gain, and a plurality of interpolation linear prediction coefficient (or line spectrum pair) candidates calculated using the linear prediction coefficients (or line spectrum pair) of the preceding and succeeding frames, and Means for selecting any one of a plurality of combinations of interpolation gain candidates calculated using the gain of the frame, and encoding by a combination of frames to be quantized by the index number. Device.