JP5084360B2 - Speech coding apparatus and speech decoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech encoding device which selects the optimal code book, even when an excitation sound source signal of a fixed sound source code book becomes a wave packet feature having a predetermined time length by impulse response of a filter placed in a post-stage of a pulse sound source code book composing the fixed sound source code book. <P>SOLUTION: An adaptive code book 8 and the fixed sound source code book 9 output not only the excitation sound source signal of a present subframe, but also the excitation sound source signal of the following subframe, and they are constituted in such a way that, not only the sound signal of the present subframe but also the sound signal of the following subframe is included for an evaluation object of a quantization error. <P>COPYRIGHT: (C)2009,JPO&amp;INPIT

Description

  The present invention relates to a speech encoding apparatus that performs high-efficiency encoding on input speech, and a speech decoding apparatus that decodes speech encoded by the speech encoding apparatus.

  As a method for highly efficient coding of a speech signal using code excitation linear prediction (CELP), a method of expressing a noise component of an excitation signal using an algebraic codebook composed of unit pulses (algebraic code excitation). There is linear prediction: Algebraic Code-Excited Linear Prediction (ACELP), and this method is adopted in various standard methods (see, for example, Non-Patent Documents 1, 2, and 3).

FIG. 31 is a block diagram showing a conventional speech coding apparatus disclosed in Non-Patent Document 2 below, for example.
The outline of the speech encoding process by the conventional speech encoding apparatus is as follows.
(1) When a speech signal as shown in FIG. 33 is input, the speech encoding apparatus divides the input speech into a certain frame length (usually about 5 msec to 50 msec, 20 msec in the AMR system).
(2) The pre-processing unit performs band-limiting filtering to remove a signal in a band not to be encoded from the input speech.
(3) The linear prediction analysis unit performs speech spectrum analysis (LPC analysis) for each frame of the input speech to calculate a linear prediction coefficient LPC to be used as a coefficient of the synthesis filter. Convert to spectrum versus LSP.

(4) The LSP quantization / inverse quantization unit performs vector quantization with reference to the LSP codebook.
That is, among the LSP coefficients recorded in the LSP codebook, the LSP coefficient closest to the line spectrum pair LSP is specified, and the index of the LSP coefficient is extracted from the LSP codebook.
Further, the LSP quantization / inverse quantization unit outputs the index of the LSP coefficient as spectrum information to the multiplexing unit.
(5) The LSP / LPC converter converts the LSP coefficient into a linear prediction coefficient LPC, and forms a synthesis filter in accordance with the linear prediction coefficient LPC.

(6) The sub-frame unit (one frame is divided into a plurality of parts on the time axis (four divisions in the AMR method, one sub-frame = 5 msec)) output from the adaptive excitation codebook and the fixed excitation codebook by the driving excitation generator A plurality of driving sound sources are generated by combining the excitation sound source signals of (section).
(7) A plurality of driving sound sources generated by the driving sound source generation unit are passed through a synthesis filter to generate a plurality of synthesized sounds.

(8) The minimum error search unit controls quantization errors between a plurality of synthesized speech and input speech while controlling the excitation excitation signal output from the adaptive codebook and the fixed excitation codebook and the gain output from the gain codebook. Evaluate and search for a synthesized speech with a minimum quantization error among a plurality of synthesized speech.
(9) Spectrum information output from the LPC quantization / inverse quantization unit by the multiplexing unit, pitch information of the excitation sound source signal output from the adaptive codebook when a synthesized speech with a minimum quantization error is obtained, The pulse information of the excitation excitation signal output from the fixed excitation codebook when the synthesized speech with the minimum quantization error is obtained, and the gain codebook output when the synthesized speech with the minimum quantization error is obtained. The gain information indicating the gain is multiplexed and the multiplexed signal is transmitted to the speech decoding apparatus.

Here, the adaptive codebook of the speech coding apparatus is an accumulation of drive excitations generated in the past.
As the fixed excitation codebook, for example, in the AMR system, as shown in FIG. 34, a pulse excitation codebook (algebraic codebook) composed of a plurality of unit pulses is used.
In addition, there is a method for improving the sound quality of the vowel part by installing a pitch emphasis filter after the pulse excitation codebook and enhancing the pitch frequency component according to the pit period of the excitation excitation signal output from the adaptive codebook. It has been taken.
The gain codebook stores a plurality of gain value candidates, and each gain value is indexed.
The driving sound source is generated by appropriately combining these codebook elements.

FIG. 32 is a block diagram showing a conventional speech decoding apparatus disclosed in Non-Patent Document 2 below, for example.
The outline of the speech decoding process by the conventional speech decoding apparatus is as follows.
(1) The demultiplexing unit receives the multiplexed signal transmitted from the speech coding apparatus, demultiplexes the multiplexed signal, and outputs spectrum information, pitch information, pulse information, and gain information.
(2) A synthesis filter is formed according to the index of the LSP coefficient indicated by the spectrum information output from the demultiplexing unit.
(3) An excitation excitation signal (adaptive codebook component signal) in the subframe corresponding to the pitch information is acquired from the adaptive codebook, and an excitation excitation signal (pulse) in the subframe corresponding to the pulse information from the fixed excitation codebook A sound source codebook component signal) is acquired.

(4) The gain indicated by the gain information is acquired from the gain codebook.
(5) The gain obtained from the gain codebook by the gain multiplier is multiplied by the adaptive codebook component signal and the pulse excitation codebook component signal, and the adder obtains the adaptive codebook component signal and the pulse excitation codebook component signal after gain multiplication. to add.
(6) Pass the excitation excitation signal (adaptive codebook component signal + pulse excitation codebook component signal) after addition by the adder through the synthesis filter to decode the synthesized speech.

The following Patent Document 1 shows an example in which the speech quality is further improved while maintaining a low bit rate by using the CELP coding scheme framework described above, and the user transmits natural and easy-to-hear speech. It is disclosed.
That is, in the following Patent Document 1, ITU-T Recommendation G. A technique using a pulse spreading codebook shown in 729 Annex D is disclosed.
FIG. 35 is a block diagram showing a fixed excitation codebook using a pulse spreading codebook. The fixed excitation codebook in FIG. 35 generates a fixed excitation vector by convolving a spreading pattern (fixed waveform) with a pulse excitation. is there.

  Further, in order to extract a signal of a desired frequency band from the pulse sound source and emphasize the signal of the frequency band, as shown in FIG. 36, a low-pass filter (LPF) or a high-pass filter (HPF) is placed after the signal. Sometimes.

As described above, if the CELP coding method is used, it is possible to reduce the bit rate in encoding the audio signal.
However, when various filters such as a diffusion filter and an HPF are installed at the subsequent stage of the pulse excitation codebook, it is generally known that a wave packet shape having a predetermined time length is obtained due to the impulse response of the filter.
FIG. 37 is an explanatory view showing a wave packet shape having a predetermined time length.
The upper part of FIG. 37 shows the output waveform of the pulse excitation codebook (the waveform corresponding to (1) in FIGS. 35 and 36), and the lower part shows the output waveform of the post filter (in FIGS. 35 and 36). 2). However, in order to make the figure easy to see, the waveform of (2) shows an envelope of the waveform.

When CELP encoding processing is executed using a wave packet having a predetermined time length as an excitation excitation signal of a fixed excitation codebook, the following problems occur.
FIG. 38 is an explanatory diagram for explaining a problem when the CELP encoding process is executed using a wave packet having a predetermined time length.
The upper part of FIG. 38 shows an ideal example in which the pulse position and wave packet corresponding to the waveform of (1) above exist in the subframe section to be encoded. At this time, since all of the wave packet components are subjected to quantization error evaluation of the fixed excitation codebook, accurate error evaluation is possible.

However, as shown in the lower part of FIG. 38, even if the pulse position is within the subframe section, the wave packet may straddle between the sub frames, and a part of the wave packet is a quantization error of the fixed excitation codebook. May not be subject to evaluation.
That is, when the pulse sound source is near the end (pulse B in the example of FIG. 38) of the subframe where the encoding process is performed (here, for convenience of description, expressed as the Nth subframe), As shown in the lower part of FIG. 38, a part of the wave packet may straddle the (N + 1) th subframe.
At this time, when encoding is performed with the frame configuration shown in FIG. 33, only the section B is actually subject to error evaluation (see the lower part of FIG. 38).

Since the error evaluation of the pulse B is performed only in a part of the wave packet (section B), there is a possibility that the pulse B has been omitted from the selection when the entire wave packet is evaluated.
Thus, a pulse (wave packet position) that is not originally selected may be selected by mistake.
In addition, since the encoding bit rate is finite, it is conceivable that a pulse to be originally selected is not selected (opportunity loss) instead of a pulse that has been selected by mistake.

Furthermore, since the impulse response component of the pulse selected in the subframe immediately before the subframe ((N−1) th subframe) is not carried over to the Nth subframe, the signal in section A in FIG. Quantization processing is performed as if there is no.
Therefore, for example, a pulse in the section A that does not need to be selected as in the case of the pulse D in FIG. 38 may be easily selected. As a result, instead of the pulse D, a pulse (for example, the pulse C) that should be originally selected is not selected (loss of opportunity).

  Since the speech decoding apparatus does not have a function to carry forward the impulse response component corresponding to the section A, the wave packet shape is collapsed, and there is a possibility that the effect of the convolved filter is reduced. It is done.

Republished 2003/071522 ITU-T Recommendation G.729, "Coding of Speech at 8kbit / s using Conjugate-Structure Algebraic-Code-Excited Linear-Prediction (CS-ACELP)" (TTC standard JT-G729, "8kbit / s CS-ACELP is used (Speech coding system) (Information and Communication Technology Committee, established in 1999) 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 26.090, "AMR speech codec; Transcoding functions", Version 4.0.0 (2001-03) ITU-T Recommendation G.722.2, "Wideband coding of speech at around 16kbit / s using Adaptive Multi-Rate Wideband (AMR-WB)" (TTC standard JT-G722.2, "Adaptive Multirate Wideband (AMR-WB) (Wideband speech coding of about 16kbit / s using ``, '' Information and Communication Technology Committee, established in 2004)

  Since the conventional speech coding apparatus is configured as described above, if the CELP coding method is used, a low bit rate can be achieved in speech signal coding. However, if various filters such as diffusion filters and HPFs are installed after the pulse excitation codebook, the excitation excitation signal of the fixed excitation codebook has a wave packet shape having a predetermined time length due to the impulse response of the filter. There is a problem that a correct codebook cannot be selected, and the quality of speech decoded by the speech decoding apparatus deteriorates.

  The present invention has been made to solve the above-described problems, and the excitation excitation signal of the fixed excitation codebook is generated by the impulse response of the filter installed at the subsequent stage of the pulse excitation codebook constituting the fixed excitation codebook. An object of the present invention is to obtain a speech coding apparatus capable of selecting an optimum codebook even when the wave packet shape has a predetermined time length.

  In the speech coding apparatus according to the present invention, the adaptive codebook and the fixed excitation codebook output not only the excitation excitation signal of the subframe but also the excitation excitation signal of the next subframe, and the parameter search means inputs the subframe. In addition to the speech, the input speech of the next subframe is included in the quantization error evaluation target.

  According to this invention, the adaptive codebook and the fixed excitation codebook output not only the excitation excitation signal of the subframe but also the excitation excitation signal of the next subframe, and the parameter search means not only the input speech of the subframe. Since the input speech of the next subframe is included in the evaluation target of the quantization error, the fixed excitation codebook of the fixed excitation codebook is determined by the impulse response of the filter installed after the pulse excitation codebook constituting the fixed excitation codebook. Even if the excitation sound source signal has a wave packet shape having a predetermined time length, there is an effect that an optimal codebook can be selected.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a speech coding apparatus according to Embodiment 1 of the present invention. In the figure, a buffer 1 is a memory for storing a speech signal as input speech.
The pre-processing unit 2 divides the audio signal stored in the buffer 1 into a predetermined frame length (usually about 5 msec to 50 msec, 20 msec in the AMR method), and performs band-limiting filtering to encode the audio signal from each frame. Perform processing to remove unnecessary low-frequency components that are not subject to conversion. For example, the preprocessing unit 2 includes a pole-zero filter having a cutoff frequency of 140 Hz.

The spectrum analysis unit 3 includes a linear prediction analysis unit 4, an LSP codebook 5, and an LSP quantization / inverse quantization unit 6. For each frame of the audio signal output from the preprocessing unit 2, an audio spectrum analysis is performed. (LPC analysis) is performed. The spectrum analysis unit 3 constitutes spectrum analysis means.
The linear prediction analysis unit 4 performs a speech spectrum analysis (LPC analysis) for each frame of the speech signal, calculates a linear prediction coefficient LPC to be used as a coefficient of the synthesis filter 16, and uses the linear prediction coefficient LPC as a line spectrum. A process of converting to LSP is executed.

The LSP codebook 5 is a codebook that records a plurality of LSP coefficients, and an index is assigned to each LSP coefficient.
The LSP quantization / inverse quantization unit 6 identifies an LSP coefficient that is closest to the line spectrum pair LSP among the LSP coefficients recorded in the LSP codebook 5, and extracts the LSP coefficient from the LSP codebook 5. While extracting an index, the process which outputs the index of the LSP coefficient to the multiplexing part 21 as spectrum information is implemented.

Under the instruction of the minimum error search unit 20, the drive excitation generator 7 is subframe units (one frame is divided into multiples on the time axis (4 in the AMR method). A process of generating a plurality of driving sound sources is performed by combining the excitation sound source signals in the divided section, 1 subframe = 5 msec). The driving sound source generator 7 constitutes a driving sound source generator.
The adaptive codebook 8 is a codebook that accumulates excitation excitation signals (adaptive codebook component signals), which are driving excitations generated in the past, and adaptation of the subframe and the next subframe indicated by the minimum error search unit 20. A codebook component signal is output.

The fixed excitation codebook 9 is a codebook in which a pulse excitation codebook (algebraic codebook) composed of a plurality of unit pulses is used in the AMR method, for example, and the subframe and the minimum error search unit 20 indicate An excitation excitation signal (pulse excitation codebook component signal) of the next subframe is output. A method of improving the sound quality of the vowel part by installing a pitch emphasis filter after the pulse excitation codebook and enhancing the pitch frequency component according to the pit period of the excitation excitation signal output from the adaptive codebook 8 May be taken.
The gain codebook 10 is a codebook storing a plurality of gains, and outputs the gain indicated by the minimum error search unit 20.

The gain multiplier 11 multiplies the excitation source signal (adaptive codebook component signal) in the subframe and the next subframe output from the adaptive codebook 8 by the gain output from the gain codebook 10.
The gain multiplier 12 multiplies the excitation source signal (pulse excitation codebook component signal) in the subframe and the next subframe output from the fixed excitation codebook 9 by the gain output from the gain codebook 10. To do.
The adder 13 adds the excitation source signal (adaptive codebook component signal) multiplied by the gain by the gain multiplier 11 and the excitation source signal (pulse excitation codebook component signal) multiplied by the gain by the gain multiplier 12. To implement.

The synthesized speech generation unit 14 forms a synthesis filter 16 according to the LSP coefficient specified by the LSP quantization / inverse quantization unit 6, and passes a plurality of driving sound sources generated by the driving sound source generation unit 7 through the synthesis filter 16. Then, a process for generating a plurality of synthesized speech is performed. The synthesized speech generation unit 14 constitutes a synthesized speech generation means.
The LSP / LPC conversion unit 15 converts the LSP coefficient specified by the LSP quantization / inverse quantization unit 6 into a linear prediction coefficient LPC, and performs a process of forming the synthesis filter 16 according to the linear prediction coefficient LPC.
The synthesis filter 16 is a filter that inputs the driving sound source generated by the driving sound source generation unit 7 and outputs the synthesized speech to the subtractor 18.

The reference vector assembly buffer 17 extracts the audio signal of the subframe and the audio signal of the next subframe from the audio signal of the frame unit output from the preprocessing unit 2, and the audio signal of the subframe and the next subframe. Is output to the subtracter 18 as a reference signal used for searching for a synthesized speech with the smallest quantization error.
The subtracter 18 performs a process of calculating a difference (quantization error) between the plurality of synthesized speech generated by the synthesized speech generating unit 14 and the reference signal output from the reference vector assembly buffer 17.

The auditory weighting filter 19 performs a process of giving auditory weighting to the quantization error calculated by the subtracter 18.
The minimum error search unit 20 outputs the excitation excitation signal output from the adaptive codebook 8 and the fixed excitation codebook 9 and the gain codebook 10 so that the perceptual weighting quantization error output from the perceptual weighting filter 19 is reduced. The process of searching the encoding parameter (pitch information, gain information, pulse information) regarding the synthetic | combination speech with the minimum quantization error is controlled among several synthetic | combination speeches by controlling the gain to be performed.
The reference vector assembly buffer 17, the subtracter 18, the perceptual weighting filter 19, and the minimum error search unit 20 constitute parameter search means.

  The multiplexing unit 21 includes the spectrum information output from the LPC quantization / inverse quantization unit 6, the pitch information of the excitation sound source signal output from the adaptive codebook 8 when the synthesized speech with the minimum quantization error is obtained, and The pulse information of the excitation excitation signal output from the fixed excitation codebook 9 when the synthesized speech with the minimum quantization error is obtained, and the output from the gain codebook 10 when the synthesized speech with the minimum quantization error is obtained. A process of multiplexing the gain information indicating the gain to be transmitted and transmitting the multiplexed signal to the speech decoding apparatus is performed.

FIG. 2 is a block diagram showing a speech decoding apparatus according to Embodiment 1 of the present invention. In FIG. 2, a demultiplexing unit 31 receives a multiplexed signal transmitted from the speech encoding apparatus, and receives the multiplexed signal. The spectrum information, pitch information, pulse information, and gain information are output. The demultiplexing unit 31 constitutes information receiving means.
The adaptive codebook 32 is a codebook corresponding to the adaptive codebook 8 in the speech coding apparatus of FIG. 1, and the excitation sound source signal (adaptive codebook component) of the subframe corresponding to the pitch information output from the demultiplexing unit 31. Signal).

The fixed excitation codebook 33 is a codebook corresponding to the fixed excitation codebook 9 in the speech encoding apparatus of FIG. 1, and the excitation excitation signal (pulse excitation source) of the subframe corresponding to the pulse information output from the demultiplexing unit 31. Codebook component signal).
The gain codebook 34 is a codebook corresponding to the gain codebook 10 in the speech coding apparatus of FIG. 1, and outputs gains corresponding to the gain information output from the demultiplexing unit 31 to the gain multipliers 35 and 36.

The gain multiplier 35 multiplies the excitation source signal (adaptive codebook component signal) in the subframe output from the adaptive codebook 32 by the gain output from the gain codebook 34.
The gain multiplier 36 performs processing for multiplying the excitation excitation signal (pulse excitation codebook component signal) in the subframe output from the fixed excitation codebook 33 by the gain output from the gain codebook 34.
The adder 37 adds the excitation excitation signal (adaptive codebook component signal) multiplied by the gain by the gain multiplier 35 and the excitation excitation signal (pulse excitation codebook component signal) multiplied by the gain by the gain multiplier 36. A process of generating a driving sound source is performed.
The adaptive codebook 32, fixed excitation codebook 33, gain codebook 34, gain multipliers 35 and 36, and adder 37 constitute drive excitation generation means.

The LSP codebook 38 is a codebook corresponding to the LSP codebook 5 in the speech coding apparatus of FIG. 1 and outputs LSP coefficients corresponding to the spectrum information output from the demultiplexing unit 31.
The LSP / LPC converter 39 converts the LSP coefficient output from the LSP codebook 38 into a linear prediction coefficient LPC, and performs a process of forming a synthesis filter 40 according to the linear prediction coefficient LPC.
The synthesis filter 40 is a filter that receives the driving sound source output from the adder 37 and outputs the synthesized speech to the post filter 41.
The post filter 41 performs processing for improving the quality of the synthesized speech output from the synthesis filter 40.
The LSP codebook 38, the LSP / LPC converter 39, the synthesis filter 40, and the post filter 41 constitute a synthesized speech decoding unit.

Next, the operation will be described.
When the buffer 1 stores an audio signal as input speech, the pre-processing unit 2 of the audio encoding device stores the audio signal in a certain frame length (usually about 5 msec to 50 msec, in the AMR system, as shown in FIG. 20 msec) and band limiting filtering is performed to remove unnecessary low-frequency components that are not to be encoded from each frame of the audio signal.

When the spectrum analysis unit 3 receives the preprocessed speech signal from the preprocessing unit 2, the spectrum analysis unit 3 performs linear prediction analysis (LPC analysis) of speech for each frame of the speech signal, and performs coefficients of the synthesis filter 16. The linear prediction coefficient (LPC) used for is calculated.
Here, the synthesis filter 16 is defined by the following equation (1).
However, A hat (z) is a coefficient of the synthesis filter 16, and a _i hat is a quantized linear prediction coefficient.

Hereinafter, the processing content of the spectrum analysis part 3 is demonstrated concretely.
The linear prediction analysis unit 4 of the spectrum analysis unit 3 performs one-line type prediction analysis for each frame using, for example, an autocorrelation method using a 30 ms wide asymmetric window.
That is, the autocorrelation coefficient of the windowed speech is calculated every 160 samples (20 ms), and the autocorrelation coefficient is converted into a linear prediction coefficient using the Levinson algorithm.
Also, the linear prediction analysis unit 4 converts the linear prediction coefficient into a line spectrum pair (LSP) so that the LSP quantization / inverse quantization unit 6 in the subsequent stage can efficiently perform quantization and interpolation of the linear prediction coefficient. Convert.

When receiving the line spectrum pair LSP from the linear prediction analysis unit 4, the LSP quantization / inverse quantization unit 6 refers to the LSP codebook 5 and performs vector quantization.
That is, the LSP quantization / inverse quantization unit 6 identifies the LSP coefficient closest to the line spectrum pair LSP among the LSP coefficients recorded in the LSP codebook 5, and Extract the LSP coefficient index.
Further, the LSP quantization / inverse quantization unit 6 outputs the index of the LSP coefficient to the multiplexing unit 21 as spectrum information.
The LSP coefficient quantized here is used in the synthesis filter 16 of the fourth subframe. Further, in order to calculate linear prediction coefficients used in the first, second, and third subframes, interpolation processing is performed using the LSP coefficients quantized in the immediately preceding frame and the quantized LSP.

The driving excitation generator 7 generates a plurality of driving excitations by combining the excitation excitation signals in units of subframes output from the adaptive codebook 8 and the fixed excitation codebook 9 under the instruction of the minimum error search unit 20.
Hereinafter, the processing content of the drive sound source generation unit 7 will be specifically described.

The adaptive codebook 8 of the drive excitation generator 7 stores excitation excitation signals (adaptive codebook component signals) that are drive excitations generated in the past, and among these excitation excitation signals (adaptive codebook component signals). To the excitation source signal (adaptive codebook component signal) instructed by the minimum error search unit 20.
The fixed excitation codebook 9 uses a pulse excitation codebook (algebraic codebook) composed of a plurality of unit pulses, and an excitation excitation signal (pulse excitation codebook) indicated by the minimum error search unit 20. Component signal).
However, the adaptive codebook 8 and the fixed excitation codebook 9 are not limited to the excitation excitation signal (adaptive codebook component signal and pulse excitation codebook component signal) of the subframe, but also the excitation excitation signal (adaptive code) over the next subframe. (Book component signal, pulse excitation codebook component signal) are also output.

  That is, in the case of the conventional speech encoding apparatus, when the current encoding target frame is the Mth frame, the excitation signal (adaptive codebook component signal) of the first subframe is used in the first subframe of the Mth frame. , Only a pulse excitation codebook component signal) is output to generate a driving excitation. As described above, a diffusion filter, an HPF, or the like is provided after the pulse excitation codebook constituting the fixed excitation codebook. When various filters are installed (see FIG. 35 and FIG. 36), the excitation excitation signal of the fixed excitation codebook has a wave packet shape having a predetermined time length due to the impulse response of the filter (see FIG. 37). Even if the quantization error between the synthesized speech generated from the excitation signal of the frame and the reference signal of the subframe is evaluated, it is not possible to select the optimum codebook A.

Therefore, in the first embodiment, the adaptive codebook 8 and the fixed excitation codebook 9 are used not only for the excitation excitation signal (adaptive codebook component signal and pulse excitation codebook component signal) of the subframe but also in the next subframe. Overlying excitation excitation signals (adaptive codebook component signal, pulse excitation codebook component signal) are also output.
Specifically, in the first subframe of the Mth frame, the excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) of the first subframe and the first subframe to the second subframe are straddled. Excitation excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) is output.
Similarly, in the Nth (N = 2, 3) subframe of the Mth frame, the excitation excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) of the Nth subframe and the (N + 1) th subframe of the Nth subframe. Excitation excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) is output.
Also, in the fourth subframe of the Mth frame, the excitation excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) of the fourth subframe of the Mth frame and the first subframe of the (M + 1) th frame. Excitation excitation signal (adaptive codebook component signal, pulse excitation codebook component signal) is output.

The gain codebook 10 is a codebook storing a plurality of gains, and outputs a gain indicated by the minimum error search unit 20 from the plurality of gains.
The gain multiplier 11 multiplies the excitation source signal (adaptive codebook component signal) in the subframe and the next subframe output from the adaptive codebook 8 by the gain output from the gain codebook 10.
The gain multiplier 12 multiplies the excitation source signal (pulse excitation codebook component signal) in the subframe and the next subframe output from the fixed excitation codebook 9 by the gain output from the gain codebook 10. .
The adder 13 adds the excitation excitation signal (adaptive codebook component signal) multiplied by the gain by the gain multiplier 11 and the excitation excitation signal (pulse excitation codebook component signal) multiplied by the gain by the gain multiplier 12. To generate a driving sound source.

When the LSP quantization / inverse quantization unit 6 specifies an LSP coefficient, the LSP / LPC conversion unit 15 of the synthesized speech generation unit 14 converts the LSP coefficient into a linear prediction coefficient LPC, and according to the linear prediction coefficient LPC A synthesis filter 16 is formed.
When the driving sound source generator 7 generates a plurality of driving sound sources, the synthesizing filter 16 of the synthesized sound generating unit 14 inputs a plurality of driving sound sources, generates a plurality of synthesized sounds from the driving sound sources, and generates a plurality of synthesized sounds. The sound is output to the subtracter 18.

When receiving the frame unit audio signal from the preprocessing unit 2, the reference vector assembly buffer 17 extracts the subframe audio signal and the next subframe audio signal from the frame unit audio signal. The audio signal of the frame and the next subframe is output to the subtracter 18 as a reference signal used for searching for synthesized speech with the minimum quantization error.
Specifically, in the first subframe of the Mth frame, as shown in FIG. 3, the audio signal of the first subframe and the audio signal of the second subframe are output as reference signals.
Similarly, in the Nth (N = 2, 3) subframe of the Mth frame, the audio signal of the Nth subframe and the audio signal of the (N + 1) th subframe are output as reference signals.
Also, in the fourth subframe of the Mth frame, the audio signal of the fourth subframe of the Mth frame and the audio signal of the first subframe of the (M + 1) th frame are output as reference signals.

The subtractor 18 calculates a difference (quantization error) between the plurality of synthesized speech generated by the synthesized speech generation unit 14 and the reference signal output from the reference vector assembly buffer 17, and uses the quantization error as the perceptual weighting filter 19. Output to.
The perceptual weighting filter 19 adds the perceptual weight to the quantization error having a flat transmission frequency response for the quantization error signal obtained by the subtractor 18 to improve the performance of the audio signal.

  The minimum error search unit 20 outputs the excitation excitation signal output from the adaptive codebook 8 and the fixed excitation codebook 9 and the gain codebook 10 so that the quantization error output from the perceptual weighting filter 19 is reduced. The synthesized speech with the smallest quantization error is searched for among the plurality of synthesized speech output from the synthesis filter 16.

  When the minimum error search unit 20 searches for the synthesized speech with the smallest quantization error, the multiplexing unit 21 searches for the spectrum information output from the LPC quantization / inverse quantization unit 6 and the synthesized speech with the smallest quantization error. Is obtained from the adaptive codebook 8 and the subframe output from the fixed excitation codebook 9 when the synthesized speech with the minimum quantization error is obtained. And the gain information indicating the gain output from the gain codebook 10 when the synthesized speech with the minimum quantization error is obtained, and the multiplexed signal is transmitted to the speech decoding apparatus. Send.

The demultiplexing unit 31 of the speech decoding apparatus receives the multiplexed signal transmitted from the speech encoding apparatus, separates the multiplexed signal, and outputs spectrum information, pitch information, pulse information, and gain information.
When adaptive codebook 32 receives pitch information from demultiplexing section 31, adaptive codebook 32 outputs an excitation excitation signal (adaptive codebook component signal) of the subframe corresponding to the pitch information.

When fixed excitation codebook 33 receives pulse information from demultiplexing unit 31, fixed excitation codebook 33 outputs an excitation excitation signal (pulse excitation codebook component signal) of the subframe corresponding to the pulse information.
When gain codebook 34 receives gain information from demultiplexing section 31, gain codebook 34 outputs gain corresponding to the gain information to gain multipliers 35 and 36.

The gain multiplier 35 multiplies the excitation source signal (adaptive codebook component signal) in the subframe output from the adaptive codebook 32 by the gain output from the gain codebook 34.
The gain multiplier 36 multiplies the excitation excitation signal (pulse excitation codebook component signal) in the subframe output from the fixed excitation codebook 33 by the gain output from the gain codebook 34.
The adder 37 adds the excitation excitation signal (adaptive codebook component signal) multiplied by the gain by the gain multiplier 35 and the excitation excitation signal (pulse excitation codebook component signal) multiplied by the gain by the gain multiplier 36. To generate a driving sound source.

When receiving the spectrum information from the demultiplexing unit 31, the LSP codebook 38 outputs LSP coefficients corresponding to the spectrum information.
When receiving the LSP coefficient from the LSP codebook 38, the LSP / LPC conversion unit 39 converts the LSP coefficient into a linear prediction coefficient LPC, and forms a synthesis filter 40 according to the linear prediction coefficient LPC.
The synthesis filter 40 receives the driving sound source output from the adder 37 and decodes the synthesized speech from the driving sound source.
When the post filter 41 receives the synthesized speech from the synthesis filter 40, the post filter 41 performs processing for improving the quality of the synthesized speech.

  As apparent from the above, according to the first embodiment, the adaptive codebook 8 and the fixed excitation codebook 9 output not only the excitation source signal of the subframe but also the excitation source signal of the next subframe, Since not only the sub-frame speech signal but also the next sub-frame speech signal is included in the quantization error evaluation target, the filter installed in the subsequent stage of the pulse excitation codebook constituting the fixed excitation codebook 9 Even if the excitation sound source signal of the fixed excitation codebook 9 has a wave packet shape having a predetermined time length due to the impulse response, the optimum codebook can be selected.

That is, according to the first embodiment, as shown in FIG. 4, even if the section where the pulse exists is limited to the section of the subframe, the section where the impulse response component extends to the next subframe (see FIG. 4). In the example, since the quantization error is evaluated until section C), a pulse (wave packet position) that is not originally selected is not erroneously selected.
In addition, although the encoding bit rate is finite, even if a pulse is selected by mistake, a pulse that should be originally selected is not selected (opportunity loss) instead of the pulse.
As described above, since the wave packet position can be arranged at a more appropriate position, the voice quality can be improved at the same bit rate.

Embodiment 2. FIG.
FIG. 5 is a block diagram showing a speech encoding apparatus according to Embodiment 2 of the present invention. In the figure, the same reference numerals as those in FIG.
Similar to the minimum error search unit 20 in FIG. 1, the weighted minimum error search unit 22 performs a process of searching for a coding parameter related to a synthesized speech having a minimum quantization error among a plurality of synthesized speech. When the error search unit 22 searches for the coding parameter related to the synthesized speech with the minimum quantization error, the error search unit 22 compares the quantization error related to the speech signal of the subframe and the quantization related to the speech signal of the next subframe. In order to relatively reduce the error evaluation, as shown in FIG. 6, the weighting for the quantization error related to the audio signal of the subframe is made larger than the weighting for the quantization error related to the audio signal of the next subframe. To do.
The reference vector assembly buffer 17, the subtracter 18, the perceptual weighting filter 19, and the weighted minimum error search unit 22 constitute parameter search means.

Next, the operation will be described.
Since the operations other than the weighted minimum error search unit 22 are the same as those in the first embodiment, only the operation of the weighted minimum error search unit 22 will be described.

For example, the coefficients of the synthesis filter 16 are usually different between the Nth subframe and the (N + 1) th subframe.
If the variation (coefficient of spectrum envelope information) between the coefficient of synthesis filter 16 in the Nth subframe and the coefficient of synthesis filter 16 in the (N + 1) th subframe is small, the method as in the first embodiment is used. However, if the fluctuation of the spectrum envelope information is large (for example, when the voice signal is noisy), quantization is performed across the (N + 1) th subframe using the spectrum envelope information of the Nth subframe. When the error is evaluated, the evaluation value is different from the original value, and a pulse to be originally selected may not be selected, or a pulse that is not originally selected may be erroneously selected.
Therefore, the second embodiment aims to absorb the fluctuation of the spectrum envelope information by relatively reducing the evaluation of the quantization error related to the audio signal of the next subframe.

When searching for a synthesized speech with the smallest quantization error, the weighted minimum error search unit 22 compares the quantization error associated with the speech signal in the next subframe as compared with the evaluation of the quantization error associated with the speech signal in the subframe. In order to reduce the evaluation relatively, the weighting for the quantization error related to the audio signal of the subframe is made larger than the weighting for the quantization error related to the audio signal of the next subframe.
Hereinafter, specific processing contents of the weighted minimum error search unit 22 will be described.

ただし、式（２）において、ベクトルｘは符号帳探索用のターゲットベクトル、ベクトルｃ_kはインデックスｋに対応する固定符号帳ベクトル、Ｈは合成フィルタ１６のインパルス応答行列である。 The error evaluation in CELP encoding is usually performed by using R (k) in the following equation (2) equivalent to the normalized inner product value of the reference signal vector and the combined signal vector for the purpose of reducing the amount of calculation. Use.
That is, the weighted minimum error search unit 22 realizes a codebook search (search for a synthesized speech with a minimum quantization error) by searching for a value of k that maximizes the error evaluation R (k).

In equation (2), vector x is a target vector for codebook search, vector _ck is a fixed codebook vector corresponding to index k, and H is an impulse response matrix of synthesis filter 16.

Here, the target vector x can be separated into x _current corresponding to the Nth subframe interval and x _next corresponding to the (N + 1) th subframe interval.
x _current
= (X ₀ , x ₁ ,..., X _M−2 , x _M−1 , 0, 0,..., 0)
(4)
x _next
= (0, 0, ..., 0, 0, x _M , x _{M + 1} , ..., x _{M + P-2} , x _{M + P-1} )
(5)

By writing the elements of the target vector x as in equations (4) and (5), the numerator component C (k) of R (k) can be expressed by the following equation corresponding to the quantization error evaluation parameter of the Nth subframe ( and C _current (k) 6) can be separated to the sum of the first (N + 1) C _next following equations corresponding to the quantization error evaluation parameter of the sub-frame (7) (k).

When the weight for the quantization error in the section after the next subframe is relatively lightened, it can be realized by multiplying C _next (k) by a weighting coefficient α as shown in the following equation (8). However, the condition of α is as shown in the following formula (9).

  As is apparent from the above, according to the second embodiment, when the weighted minimum error search unit 22 searches for the synthesized speech with the minimum quantization error, the weighting for the quantization error related to the speech signal of the subframe is performed. Is set to be larger than the weighting for the quantization error related to the audio signal of the next subframe, so that it is possible to maintain good audio quality even if the spectrum envelope fluctuates rapidly. Further, the same effect can be obtained by evaluating the coefficient fluctuation range of the synthesis filter 16 and adaptively determining the value of α.

Embodiment 3 FIG.
In Embodiments 1 and 2 described above, not only the audio signal of the subframe but also the audio signal of the next subframe is included in the quantization error evaluation target, but the next subframe is different from the subframe. May belong to the quantization error evaluation target.
Here, FIG. 7 is an explanatory diagram showing the waveform of the input sound, the frame and subframe of the input sound, the reference signal, and the like.
The error evaluation interval for the first to third subframes is the same as in the first embodiment (see FIG. 3), but the error evaluation interval for the fourth subframe is limited to the fourth subframe interval itself. This is different from the first embodiment.

In the second embodiment, the method for avoiding erroneous selection of pulses due to fluctuations in the spectral envelope has been described. However, the coefficients of the synthesis filter between subframes in the same frame are obtained by linear interpolation of the synthesis filter coefficients between frames. Therefore, the fluctuation range is considered to be relatively moderate.
However, since the frame is recalculated by performing linear prediction analysis again from each speech signal, the fluctuation range becomes larger.
Therefore, in this third embodiment, by not performing the quantization error evaluation in the section across the frames, it has a higher tolerance, and even if the spectrum envelope fluctuates rapidly, a good voice quality is obtained. So that it can be maintained.

Embodiment 4 FIG.
8 is a block diagram showing a speech encoding apparatus according to Embodiment 4 of the present invention. In the figure, the same reference numerals as those in FIG.
The carry component storage memory 23 is a memory for storing the impulse response component of the fixed excitation codebook 9 in the previous subframe.
The adder 24 adds the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23 to the excitation excitation signal of the subframe output from the fixed excitation codebook 9. To implement.
The carry component storage memory 23 and the adder 24 constitute an adding means.

FIG. 9 is a block diagram showing a speech decoding apparatus according to Embodiment 4 of the present invention. In the figure, the same reference numerals as those in FIG.
The carry component storage memory 42 is a memory for storing the impulse response component of the fixed excitation codebook 33 in the previous subframe.
The adder 43 adds the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in the carry component storage memory 42 to the excitation excitation signal of the subframe output from the fixed excitation codebook 33. To implement.
The carry component storage memory 42 and the adder 43 constitute an adding means.

Next, the operation will be described.
In the fourth embodiment, the impulse response component of the fixed excitation codebook 33 in the previous subframe is added to the excitation excitation signal of the subframe output from the fixed excitation codebook 33. This is different from Form 1.

In the speech coding apparatus, the carry component storage memory 23 stores the impulse response component of the fixed excitation codebook 9 in the previous subframe, and the adder 24 outputs the excitation excitation signal of the subframe output from the fixed excitation codebook 9. In addition, the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23 is added.
For example, in the case of the pulse A selected in the (N−1) th subframe (see FIG. 10), the impulse response component (section A = C _{previous in} FIG. 10) carried over to the Nth subframe is the (N -1) Since the pulse A is determined when the subframe is processed, the carry component storage memory 23 stores the signal C _previous .

ただし、Ｃ_previousは、第（Ｎ−１）サブフレームにおける固定音源符号帳９のインパルス応答成分であり、繰越成分記憶用メモリ２３に記憶される信号Ｃ_previousである。 When the processing shifts from the (N−1) th subframe to the Nth subframe, the adder 24 extracts the signal C _previous from the carry component storage memory 23 and outputs the signal C _previous from the fixed excitation codebook 9. It adds to the excitation excitation signal of the said sub-frame, and uses the addition result for the search of a fixed excitation codebook.
In the fourth embodiment, when the minimum error search unit 20 searches for a value of k that maximizes the error evaluation R (k) of Equation (2), the molecular component C (k) of the error evaluation R (k) is The following formula (10) is used.

Here, C _previous is an impulse response component of the fixed excitation codebook 9 in the (N−1) th subframe, and is a signal C _previous stored in the carry component storage memory 23.

In the speech decoding apparatus, in the processing of the (N−1) th subframe, pulse information indicating the position of the pulse A is transmitted from the speech encoding apparatus, but the fixed excitation codebook 33 is fixed to the speech encoding apparatus. Since it has the same filter as the excitation codebook 9 (impulse response information of the same internal filter as the fixed excitation codebook 9), an impulse response component that is carried forward to the Nth subframe can be automatically obtained. The storage memory 42 stores the signal C _previous which is an impulse response component.
When the process proceeds from the (N−1) th subframe to the Nth subframe, the adder 43 extracts the signal C _previous from the carry component storage memory 42 and outputs the signal C _previous from the fixed excitation codebook 33. The result is added to the excitation sound source signal of the subframe, and the addition result is output to the gain multiplier 36.

As is apparent from the above, according to the fourth embodiment, the impulse response of the fixed excitation codebooks 9 and 33 in the previous subframe is added to the excitation excitation signal of the relevant subframe output from the fixed excitation codebooks 9 and 33. Since the components are added, the sound quality can be further improved as compared with the first embodiment.
For example, as in the pulse D in FIG. 10, it is difficult to select a pulse in the section A that does not need to be selected, and a pulse to be originally selected (for example, the pulse C) is selected. Contributes to improving sound quality.
In the speech decoding apparatus, since the impulse response component corresponding to the section A is provided with the function of carrying over and reproducing, the wave packet shape is not lost, and the adverse effects such as reducing the effect of the convoluted filter are not generated. Quality is improved.

Embodiment 5 FIG.
FIG. 11 is a block diagram showing a part of a speech encoding apparatus according to Embodiment 5 of the present invention. In the figure, the same reference numerals as those in FIG.
The gain multiplier 25 receives a fixed gain of 0 or more and less than 1 in the signal C _previous which is the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23, or the passage of time. Accordingly, a process of multiplying the gain gradually decreasing from 1 to 0 and outputting the signal C _previous after gain multiplication to the adder 24 is performed. The gain multiplier 25 constitutes addition means.

FIG. 12 is a block diagram showing a part of a speech decoding apparatus according to Embodiment 5 of the present invention. In FIG. 12, the same reference numerals as those in FIG.
The gain multiplier 44 has a fixed gain of 0 or more and less than 1 in the signal C _previous which is the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 42, or the passage of time. Accordingly, a process of multiplying the gain gradually decreasing from 1 to 0 and outputting the signal C _previous after gain multiplication to the adder 24 is performed. The gain multiplier 44 constitutes addition means.

In Embodiment 4 described above, the impulse response component of fixed excitation codebooks 9 and 33 in the previous subframe is added to the excitation excitation signal of the subframe output from fixed excitation codebooks 9 and 33. The gain multipliers 25 and 44 of the speech encoding device and speech decoding device multiply the signal C _previous which is the impulse response component of the fixed excitation codebooks 9 and 33 in the previous subframe by a gain of 0 or more and less than 1. The weighting of the impulse response components of the fixed excitation codebooks 9 and 33 in the previous subframe may be reduced.

By adopting such a configuration, it becomes possible to withstand a sudden fluctuation in the spectrum envelope.
Also, when decoding synthesized speech, the weight of the carry-over component from the previous subframe becomes relatively light, so that even if the previous subframe is lost and the carry-over component of the subframe is lost, damage is caused. It becomes lighter and can withstand frame loss.

Embodiment 6 FIG.
FIG. 13 is a block diagram showing a part of a speech encoding apparatus according to Embodiment 6 of the present invention. In the figure, the same reference numerals as those in FIG.
If the previous subframe belongs to the same frame as the subframe, the switch 26 adds the signal C _previous which is the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23. However, when the previous sub-frame belongs to a different frame from the sub-frame, the signal C _previous is not provided to the adder 24. Note that the switch 26 constitutes an adding means.

FIG. 14 is a block diagram showing a part of a speech decoding apparatus according to Embodiment 6 of the present invention. In the figure, the same reference numerals as those in FIG.
If the previous subframe belongs to the same frame as the subframe, the switch 45 adds the signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in the carry component storage memory 42. However, when the previous subframe belongs to a different frame from the subframe, the signal C _previous is not provided to the adder 43. Note that the switch 45 constitutes an adding means.

  In Embodiment 4 described above, the impulse response component of fixed excitation codebooks 9 and 33 in the previous subframe is added to the excitation excitation signal of the subframe output from fixed excitation codebooks 9 and 33. In the sixth embodiment, when the previous subframe belongs to a frame different from the subframe, the excitation signal of the subframe output from the fixed excitation codebooks 9 and 33 is added to the fixed excitation in the previous subframe. The impulse response components of the codebooks 9 and 33 are not added.

That is, as shown in FIG. 15, the switches 26 and 45 of the speech encoding device and speech decoding device turn on the switch state and fix in the previous subframe if the subframe is the second to fourth subframes. The signal C _previous which is an impulse response component of the excitation codebook 33 is _supplied to the adder 43. If the subframe is the first subframe immediately after the previous frame, the switch state is turned OFF, The signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the subframe is not given to the adder 43.

By adopting such a configuration, it becomes possible to withstand a sudden fluctuation in the spectrum envelope.
Also, when decoding synthesized speech, the weight of the carry-over component from the previous subframe becomes relatively light, so that even if the previous subframe is lost and the carry-over component of the subframe is lost, damage is caused. It becomes lighter and can withstand frame loss.

In the sixth embodiment, the switches 26 and 45 are turned on and off to control the output of the signal C _previous to the adders 24 and 43. However, instead of the switches 26 and 45, 11 and 12 may be used, and the gains of the gain multipliers 25 and 44 may be switched to 0 or 1.

Embodiment 7 FIG.
FIG. 16 is a block diagram showing a speech encoding apparatus according to Embodiment 7 of the present invention. In the figure, the same reference numerals as those in FIG.
Similar to the minimum error search unit 20 in FIGS. 1 and 8, the mode selection function-equipped minimum error search unit 27 reduces the quantization error output from the perceptual weighting filter 19 and reduces the adaptive codebook 8 and the fixed excitation code. The excitation sound source signal output from the book 9 and the gain output from the gain codebook 10 are controlled, and among the plurality of synthesized voices output from the synthesis filter 16, the synthesized voice with the minimum quantization error is related. In addition to performing the process of searching for the coding parameter, it is determined whether or not to add the impulse response component of the fixed excitation codebook 9 in the previous subframe based on the evaluation result of the quantization error, and the determination result is Processing for outputting the mode information shown to the switch 28 and the multiplexing unit 29 is performed. The mode selection function-equipped minimum error search unit 27 constitutes parameter search means.

If the switch 28 indicates that the mode information output from the mode selection function-equipped minimum error search unit 27 is to be added, the switch state is turned ON and the previous subframe stored in the carry component storage memory 23 is turned on. A process of giving the signal C _previous which is an impulse response component of the fixed excitation codebook 9 to the adder 24 is performed. Note that the switch 28 constitutes an adding means.
The multiplexing unit 29 multiplexes spectrum information, pitch information, pulse information, and gain information as well as the multiplexing unit 21 of FIGS. 1 and 8 and also outputs the mode output from the mode selection function-equipped minimum error search unit 27. Information is also multiplexed together, and the multiplexed signal is transmitted to the speech decoding apparatus.
FIG. 18 is a flowchart showing an outline of the processing contents of the speech coding apparatus according to Embodiment 7 of the present invention.

FIG. 17 is a block diagram showing a speech decoding apparatus according to Embodiment 7 of the present invention. In the figure, the same reference numerals as those in FIG.
The demultiplexing unit 46 receives the multiplexed signal transmitted from the speech coding apparatus, demultiplexes the multiplexed signal, and outputs spectrum information, pitch information, pulse information, gain information, and mode information. The demultiplexing unit 46 constitutes information receiving means.
If the mode information output from the demultiplexer 46 indicates that the impulse response component of the fixed excitation codebook in the previous subframe is added, the switch 47 stores the previous subframe stored in the carry component storage memory 42. The signal C _previous which is the impulse response component of the fixed excitation codebook 33 in FIG. The switch 47 constitutes addition means.

In Embodiment 4 described above, the impulse response component of fixed excitation codebooks 9 and 33 in the previous subframe is added to the excitation excitation signal of the subframe output from fixed excitation codebooks 9 and 33. Based on the evaluation result of the quantization error, it may be determined whether or not to add the impulse response components of the fixed excitation codebooks 9 and 33 in the previous subframe.
Specifically, it is as follows.

The minimum error search unit 27 with a mode selection function of the speech encoding device calculates a signal C _previous that is an impulse response component of the fixed excitation codebook 42 in the previous subframe when searching for synthesized speech with the minimum quantization error. (Step ST1).
When calculating the signal C _previous which is an impulse response component, the mode selection function-equipped minimum error search unit 27 compares the signal C _previous with a predetermined threshold (step ST2), and the signal C _previous is larger than the predetermined threshold. Since the contribution of the carry-over component is large, it is determined that the impulse response component is added, and mode information indicating that the impulse response component is added is output to the switch 28 and the multiplexing unit 21 (step ST3).
Further, the molecular component C (k) of the error evaluation R (k) is calculated using the equation (10) (step ST4).

When the signal C _previous is smaller than the predetermined threshold value, the mode selection function-equipped minimum error search unit 27 does not add the impulse response component because the contribution of the carry-over component is small and the carry-over component may be a cause of deterioration. The mode information indicating that the impulse response component is not added is output to the switch 28 and the multiplexing unit 21 (step ST5).
Also, assuming that C _current (k) in equation (6) corresponding to the quantization error evaluation parameter of the subframe is the molecular component C (k) of error evaluation R (k), C _{current in} equation (6) (K) is calculated (step ST6).

If the switch 28 of the speech encoding device indicates that the mode information output from the mode selection function-added minimum error search unit 27 is to be added, the switch state is turned ON and stored in the carry component storage memory 23. If the signal C _previous which is an impulse response component of the fixed excitation codebook 9 in the previous subframe is given to the adder 24 to indicate that the mode information is not added, the switch state is turned OFF and the previous subframe is turned off. The signal C _previous which is the impulse response component of the fixed excitation codebook 9 is not given to the adder 24.

Next, the mode selection function-equipped minimum error search unit 27 calculates the error evaluation R (k) for all k using Equation (2) (step ST7), and maximizes the error evaluation R (k). The value of k to be searched is searched (step ST8).
When the minimum error search unit 27 with mode selection function searches for a value of k that maximizes the error evaluation R (k), an index (pitch information, pulse information, gain information) corresponding to the value of k is multiplexed by the multiplexing unit 29. (Step ST9).
The multiplexing unit 29 multiplexes the spectrum information, pitch information, pulse information, and gain information, and also multiplexes the mode information output from the mode selection function-equipped minimum error search unit 27, and the multiplexed signal is voiced. Send to decryption device.

The demultiplexing unit 46 of the speech decoding apparatus receives the multiplexed signal transmitted from the speech encoding apparatus, separates the multiplexed signal, and outputs spectrum information, pitch information, pulse information, gain information, and mode information. To do.
If the mode information output from the demultiplexing unit 46 indicates that the impulse response component of the fixed excitation codebook in the previous subframe is to be added, the switch 47 turns on the switch state and carries the carry component storage memory 42. If the signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in FIG. 4 is given to the adder 43 and the mode information indicates that the addition is not performed, the switch state is turned OFF, The signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe is not given to the adder 43.

  According to the seventh embodiment, since it is determined whether or not the carry-over component is added based on the result of evaluating the contribution degree of the carry-over component, the transmission bit rate is slightly increased by the amount of transmission of the mode information. Although it increases, the quantization error of the fixed excitation codebooks 9 and 33 can be effectively reduced, and there is an effect that the voice quality can be improved.

Embodiment 8 FIG.
FIG. 19 is a block diagram showing a part of a speech encoding apparatus according to Embodiment 8 of the present invention. In the figure, the same reference numerals as those in FIG.
The pitch stability evaluation unit 61 monitors the variation of the pitch period in the past subframe and performs a process of adjusting the filter coefficient of the pitch enhancement filter 63 according to the variation of the pitch period.
The pitch enhancement filter 62 is a filter in which a filter coefficient corresponding to the pitch period of the excitation excitation signal output from the adaptive codebook 8 is set and the pitch frequency component of the excitation excitation signal output from the fixed excitation codebook 9 is enhanced. .
The pitch enhancement filter 63 is a filter for which the filter coefficient is set by the pitch stability evaluation unit 61 and emphasizes the pitch frequency component of the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23. It is.
Note that the pitch stability evaluation unit 61 and the pitch emphasis filters 62 and 63 constitute addition means.

20 is a block diagram showing a part of a speech decoding apparatus according to Embodiment 8 of the present invention. In the figure, the same reference numerals as those in FIG.
The pitch stability evaluation unit 81 monitors the variation of the pitch period in the past subframe, and performs a process of adjusting the filter coefficient of the pitch enhancement filter 83 according to the variation of the pitch period.
The pitch enhancement filter 82 is a filter in which a filter coefficient corresponding to the pitch period of the excitation excitation signal output from the adaptive codebook 32 is set and the pitch frequency component of the excitation excitation signal output from the fixed excitation codebook 33 is enhanced. .
The pitch enhancement filter 83 is a filter for which the filter coefficient is set by the pitch stability evaluation unit 81 and emphasizes the pitch frequency component of the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in the carry component storage memory 42. It is.
Note that the pitch stability evaluation unit 81 and the pitch emphasis filters 82 and 83 constitute addition means.

FIG. 21 is a block diagram showing an internal configuration example of the pitch enhancement filter 63 of FIG. In the example of FIG. 21, the pitch enhancement filter 63 includes an adder 63a, an inverse filter 63b, and a filter coefficient multiplier 63c.
FIG. 21 shows an example of the internal configuration of the pitch enhancement filter 63 of FIG. 19, but the internal configuration of the pitch enhancement filter 83 of FIG. 20 is the same.

Next, the operation will be described.
The pitch stability evaluation units 61 and 81 of the speech encoding device and speech decoding device monitor changes in pitch period in past subframes.
That is, the pitch stability evaluation units 61 and 81 refer to the pitch information output from the adaptive codebook 8 or the demultiplexing unit 31 and calculate, for example, the average and variance of pitch periods in the past N subframes.

When the variance of the pitch period is small (when the pitch fluctuation is small and stable), the pitch stability evaluation units 61 and 81 increase the filter coefficients of the pitch enhancement filters 63 and 83 to increase the pitch enhancement filters 63 and 83. Try to strengthen the filter effect.
On the other hand, when the variance of the pitch period is large (when the pitch fluctuation is large and unstable), the filter coefficients of the pitch enhancement filters 63 and 83 are reduced, and the filter effect of the pitch enhancement filters 63 and 83 is invalidated, or Reduce the filter effect.

  According to the eighth embodiment, the pitch stability evaluation units 61 and 81 monitor the fluctuation of the pitch period in the past subframe and adjust the filter coefficients of the pitch enhancement filters 63 and 83 according to the fluctuation of the pitch period. Therefore, the speech encoding apparatus does not transmit mode information to the speech decoding apparatus, and the quantization error of the fixed excitation codebook is reduced while maintaining the transmission bit rate, thereby improving speech quality. There is an effect that can be.

Embodiment 9 FIG.
FIG. 22 is a block diagram showing a speech encoding apparatus according to Embodiment 9 of the present invention. In the figure, the same reference numerals as those in FIGS.
Similar to the minimum error search unit 20 in FIGS. 1 and 8, the minimum error search unit with mode selection function 64 is adapted to the adaptive codebook 8 and the fixed excitation code so that the quantization error output from the perceptual weighting filter 19 is reduced. The excitation sound source signal output from the book 9 and the gain output from the gain codebook 10 are controlled, and among the plurality of synthesized voices output from the synthesis filter 16, the synthesized voice with the minimum quantization error is related. The processing for searching for the encoding parameters is performed, but when controlling the excitation excitation signal output from the adaptive codebook 8 and the fixed excitation codebook 9, the filter coefficient of the pitch enhancement filter 84 is also controlled. The mode selection function-equipped minimum error search unit 64 constitutes parameter search means.
The pitch enhancement filter 65 has a filter coefficient set by the mode selection function-equipped minimum error search unit 64, and calculates the pitch frequency component of the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23. It is a filter to emphasize. Note that the pitch emphasis filter 65 constitutes addition means.

FIG. 23 is a block diagram showing a speech decoding apparatus according to Embodiment 9 of the present invention. In the figure, the same reference numerals as those in FIGS.
The pitch enhancement filter 84 is set with a filter coefficient corresponding to the gain index indicated by the mode information output from the demultiplexing unit 46, and the impulse of the fixed excitation codebook 33 in the previous subframe stored in the carry component storage memory 42. It is a filter that emphasizes the pitch frequency component of the response component. The pitch emphasizing filter 84 constitutes addition means.

FIG. 24 is a block diagram showing an example of the internal configuration of the pitch enhancement filter 65 of FIG. In the example of FIG. 24, the pitch enhancement filter 65 includes an adder 65a, an inverse filter 65b, a gain codebook 65c, and a filter coefficient multiplier 65d.
24 shows an example of the internal configuration of the pitch enhancement filter 65 of FIG. 12, the internal configuration of the pitch enhancement filter 84 of FIG. 23 is the same.

In the first embodiment, the minimum error search unit 20 makes the excitation excitation signal output from the adaptive codebook 8 and the fixed excitation codebook 9 so that the quantization error output from the perceptual weighting filter 19 is reduced, Although the gain output from the gain codebook 10 is controlled to search for the synthesized speech with the minimum quantization error among the plurality of synthesized speech output from the synthesis filter 16, the pitch emphasis is further shown. A synthesized speech with the smallest quantization error may be searched for while controlling the filter coefficient of the filter 84.
Specifically, it is as follows.

The mode selection function-equipped minimum error search unit 64 includes an excitation excitation signal output from the adaptive codebook 8, an excitation excitation signal output from the fixed excitation codebook 9, a gain output from the gain codebook 10, and a pitch. A plurality of driving sound sources are generated by appropriately combining the filter coefficients of the enhancement filter 84.
The minimum error search unit 64 with a mode selection function passes through the synthesis filter 16 through the plurality of driving sound sources generated by the driving sound source generation unit 5, and among the plurality of synthesized speech output from the synthesis filter 16, A synthesized speech with the smallest quantization error is searched.

When the minimum error search unit with mode selection function 64 searches for a synthesized speech with the smallest quantization error, the mode indicating a gain index (index corresponding to the filter coefficient) of the pitch enhancement filter 65 when the synthesized speech is obtained. Information is output to the multiplexing unit 29.
The multiplexing unit 29 multiplexes the spectrum information, pitch information, pulse information, and gain information, and also multiplexes the mode information output from the mode selection function-equipped minimum error search unit 64, and the multiplexed signal is voiced. Send to decryption device.

The demultiplexing unit 46 of the speech decoding apparatus receives the multiplexed signal transmitted from the speech encoding apparatus, separates the multiplexed signal, and outputs spectrum information, pitch information, pulse information, gain information, and mode information. To do.
When the pitch enhancement filter 84 receives the mode information output from the demultiplexing unit 46, the filter coefficient corresponding to the gain index indicated by the mode information is set, and the previous subframe stored in the carry component storage memory 42 is set. The pitch frequency component of the impulse response component of the fixed excitation codebook 33 is emphasized.

  According to the ninth embodiment, since it is configured to search for the synthesized speech with the minimum quantization error while controlling the filter coefficient of the pitch enhancement filter 84, the mode information indicating the gain index of the pitch enhancement filter 65 is obtained. Although the transmission bit rate is slightly increased by the amount of transmission, the quantization error of the fixed excitation codebooks 9 and 33 can be effectively reduced, and the voice quality can be improved.

Embodiment 10 FIG.
25 is a block diagram showing a speech coding apparatus according to Embodiment 10 of the present invention. In the figure, the same reference numerals as those in FIG.
Similar to the minimum error search unit 20 in FIGS. 1 and 8, the mode selection function-equipped minimum error search unit 66 reduces the quantization error output from the perceptual weighting filter 19 and reduces the adaptive codebook 8 and the fixed excitation code. The excitation sound source signal output from the book 9 and the gain output from the gain codebook 10 are controlled, and among the plurality of synthesized voices output from the synthesis filter 16, the synthesized voice with the minimum quantization error is related. In addition to performing the process of searching for the encoding parameter, it is determined whether to include not only the input speech of the subframe but also the input speech of the next subframe in the quantization error evaluation target, and the input speech of the next subframe Is included in the quantization error evaluation target, the mode information indicating that the impulse response component of the fixed excitation codebook in the previous subframe is added is added to the switch 28 and the multiplexing unit 29. And it carries out a process of outputting. The mode selection function-equipped minimum error search unit 66 constitutes a parameter search means.

FIG. 26 is a flowchart showing an outline of the processing contents of the speech coding apparatus according to Embodiment 10 of the present invention.
The configuration of the speech decoding apparatus according to Embodiment 10 of the present invention is the same as that of the speech decoding apparatus of FIG. 17, for example.

Next, the operation will be described.
The mode selection function-equipped minimum error search unit 66 determines whether or not the previous subframe is the “carry forward mode” (step ST11).
That is, in the previous subframe, the switch state of the switch 28 is turned ON, and the signal C _previous which is the impulse response component of the fixed excitation codebook 9 stored in the carry component storage memory 23 is given to the adder 24. It is determined whether or not.

When the previous subframe is the “carry forward mode”, the mode selection function-equipped minimum error search unit 66 calculates the signal C _previous stored in the carry component storage memory 23 using the above equation (11). (Step ST12), the signal C _previous is added to the numerator component C (k) of the error evaluation R (k) (see equation (10)).
On the other hand, when the previous subframe is not the “carry-over mode”, the signal C _previous = 0 (step ST13), and the signal C _previous is not added to the numerator component C (k) of the error evaluation R (k).

Next, the mode selection function-equipped minimum error search unit 66 calculates C _next (k) of Expression (7) corresponding to the quantization error evaluation parameter of the next subframe (step ST14), and C _next (k) Is compared with a predetermined threshold value (step ST15).
When C _next (k) is larger than a predetermined threshold, the mode selection function-equipped minimum error search unit 66 not only inputs the input sound of the next subframe but also the input sound of the next subframe, as in the first embodiment. The mode information indicating that the impulse response component of the fixed excitation codebook in the previous subframe is added is output to the switch 28 and the multiplexing unit 29 (step ST16).
Further, the minimum error search unit with mode selection function 66 calculates the molecular component C (k) of the error evaluation R (k) as shown in the following equation (12) (step ST17).
C (k) = C _previous + C _current (k) + C _next (k) (12)

When C _next (k) is smaller than a predetermined threshold, the mode selection function-equipped minimum error search unit 66 limits the evaluation target to only the input speech of the subframe without expanding the evaluation target of the quantization error. In addition, mode information indicating that the impulse response component of the fixed excitation codebook in the previous subframe is not added is output to the switch 28 and the multiplexing unit 29 (step ST18).
Further, the mode selection function-equipped minimum error search unit 66 calculates the numerator component C (k) of the error evaluation R (k) using the equation (10) (step ST19).

If the switch 28 of the speech encoding device indicates that the mode information output from the mode selection function-added minimum error search unit 66 is to be added, the switch state is turned ON and stored in the carry component storage memory 23. If the signal C _previous which is an impulse response component of the fixed excitation codebook 9 in the previous subframe is given to the adder 24 to indicate that the mode information is not added, the switch state is turned OFF and the previous subframe is turned off. The signal C _previous which is the impulse response component of the fixed excitation codebook 9 is not given to the adder 24.

Next, the mode selection function-equipped minimum error search unit 66 calculates the error evaluation R (k) using the equation (2) (step ST20), and among all k, the error evaluation R (k) The value of k that maximizes is searched (step ST21).
When searching for the value of k that maximizes the error evaluation R (k), the mode selection function-equipped minimum error search unit 66 multiplexes an index (pitch information, pulse information, gain information) corresponding to the value of k. (Step ST22).
The multiplexing unit 29 multiplexes the spectrum information, pitch information, pulse information, and gain information, and also multiplexes the mode information output from the mode selection function-equipped minimum error search unit 66, and the multiplexed signal is voiced. Send to decryption device.

17 receives the multiplexed signal transmitted from the speech coding apparatus, separates the multiplexed signal, and obtains spectrum information, pitch information, pulse information, gain information, and mode. Output information.
If the mode information output from the demultiplexing unit 46 indicates that the impulse response component of the fixed excitation codebook in the previous subframe is to be added, the switch 47 turns on the switch state and carries the carry component storage memory 42. If the signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in FIG. 4 is given to the adder 43 and the mode information indicates that the addition is not performed, the switch state is turned OFF, The signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe is not given to the adder 43.

  According to the tenth embodiment, since it is configured to determine whether to carry over to the subframe immediately after the subframe ((N + 1) th subframe) according to the contribution of the extended section of error evaluation, There is no mismatch between the availability of expansion and the availability of carry-over, and the sound quality can be improved.

Embodiment 11 FIG.
FIG. 27 is a block diagram showing a speech encoding apparatus according to Embodiment 11 of the present invention. In the figure, the same reference numerals as those in FIG.
As with the minimum error search unit with mode selection function 66 in FIG. 25, the mode-specific minimum error search unit 67 includes not only the input speech of the subframe but also the input speech of the next subframe as the quantization error evaluation target. If the input speech of the next subframe is included in the quantization error evaluation target, mode information indicating that the impulse response component of the fixed excitation codebook in the previous subframe is added is output to the switch 28. However, unlike the minimum error search unit 66 with a mode selection function in FIG. 25, the mode information output to the switch 28 is the switch 28 and carry-over component error evaluation as mode information candidates (mode information that is not finally determined). It outputs to the mode determination part 71.

The multiplexing unit 68 includes the spectrum information output from the LPC quantization / inverse quantization unit 6, the pitch information of the excitation sound source signal output from the adaptive codebook 8 when the synthesized speech with the minimum quantization error is obtained, and Then, a process of multiplexing the gain information indicating the gain output from the gain codebook 10 when the synthesized speech with the minimum quantization error is obtained and outputting the multiplexed signal to the buffer 69 is performed.
The buffer 69 is a memory that temporarily stores the multiplexed signal output from the multiplexing unit 68.
The buffer 70 temporarily stores the mode information candidates output from the mode-specific minimum error search unit 67, and the excitation source signal output from the fixed excitation codebook 9 when a synthesized speech with the minimum quantization error is obtained. Is temporarily stored as pulse information candidates (pulse information that has not been finally determined).

Based on the evaluation result of the quantization error in the next subframe, the carry-over component error evaluation mode determination unit 71 determines whether or not the input speech of the next subframe is included in the quantization error evaluation target, and the pulse information and A process for finally determining the mode information is performed.
The mode-specific minimum error search unit 67, multiplexing unit 68, buffers 69 and 70, and carry-over component error evaluation mode determination unit 71 constitute parameter search means.
The multiplexing unit 72 multiplexes the pulse information and mode information determined by the carry component error evaluation mode determination unit 71 and the multiplexed signal output from the buffer 69, and transmits the multiplexed signal to the speech decoding apparatus. Perform the process.

FIG. 28 is a flowchart showing an outline of the processing contents of the speech coding apparatus according to Embodiment 11 of the present invention.
The configuration of the speech decoding apparatus according to Embodiment 11 of the present invention is the same as, for example, the speech decoding apparatus of FIG.

Next, the operation will be described.
First, processing in the Nth subframe will be described.
The mode-specific minimum error search unit 67 determines whether or not the (N−1) -th subframe that is the previous subframe is the “carry-over mode” (step ST31).
That is, in the previous subframe, the switch state of the switch 28 is turned ON, and the signal C _previous which is the impulse response component of the fixed excitation codebook 9 stored in the carry component storage memory 23 is given to the adder 24. It is determined whether or not.

When the previous subframe is the “carry forward mode”, the mode-specific minimum error search unit 67 calculates the signal C _previous stored in the carry component storage memory 23 using the above equation (11) (step S1). ST32), the signal C _previous is added to the numerator component C (k) of the error evaluation R (k) (see equation (10)).
On the other hand, when the previous subframe is not the “carry-over mode”, the signal C _previous = 0 (step ST33), and the signal C _previous is not added to the numerator component C (k) of the error evaluation R (k).

Next, the mode-specific minimum error search unit 67 calculates C _next (k) in Expression (7) corresponding to the quantization error evaluation parameter of the (N + 1) th subframe which is the next subframe (step ST34).
When the mode-specific minimum error search unit 67 calculates C _next (k) in Expression (7), it outputs a mode information candidate indicating that the impulse response component of the fixed excitation codebook in the previous subframe is added to the switch 28. At the same time, the mode information candidates are stored in the buffer 70 (step ST35), and the C _next (k) is substituted into the equation (12) to calculate the molecular component C (k) of the error evaluation R (k) ( Step ST36).
When the switch 28 receives the mode information candidate indicating addition from the mode-specific minimum error search unit 67, the switch 28 turns on the fixed excitation codebook in the previous subframe stored in the carry-over component storage memory 23. The signal C _previous which is the impulse response component of 9 is _supplied to the adder 24.

Next, the mode-specific minimum error search unit 67 calculates the error evaluation R (k) using the equation (2) (step ST37), and the error evaluation R (k) is maximized among all k. The value of k to be searched is searched (step ST38).
When searching for the value of k that maximizes the error evaluation R (k), the mode-specific minimum error search unit 67 stores the value k as k0 and stores the index k0 (step ST39).
The pulse information of the excitation excitation signal output from the fixed excitation codebook 9 when the error evaluation R (k) is maximized is stored in the buffer 70 as a pulse information candidate.

Next, the mode-specific minimum error search unit 67 outputs a mode information candidate indicating that the impulse response component of the fixed excitation codebook in the previous subframe is not added to the switch 28 and stores the mode information candidate in the buffer 70. Then, the C _next (k) is substituted into the equation (10) to calculate the molecular component C (k) of the error evaluation R (k) (step ST41).
When the switch 28 receives the mode information candidate indicating that the addition is not performed from the mode-specific minimum error search unit 67, the switch 28 turns off the switch state, and the fixed excitation codebook in the previous subframe stored in the carry component storage memory 23. The signal C _previous which is the impulse response component of 9 is not given to the adder 24.

Next, the mode-specific minimum error search unit 67 calculates the error evaluation R (k) using Equation (2) (step ST42), and the error evaluation R (k) is maximized among all k. The value of k to be searched is searched (step ST43).
When searching for the value of k that maximizes the error evaluation R (k), the mode-specific minimum error search unit 67 stores the index k1 with the value of k as k1 (step ST44).
The pulse information of the excitation excitation signal output from the fixed excitation codebook 9 when the error evaluation R (k) is maximized is stored in the buffer 70 as a pulse information candidate.

Next, processing in the (N + 1) th subframe will be described.
The mode-specific minimum error search unit 67 calculates the signal C _previous stored in the carry component storage memory 23 (step ST51), and compares the signal C _previous with a predetermined threshold (step ST52).
When the signal C _previous is larger than a predetermined threshold, the mode-specific minimum error search unit 67 sets the NSF mode information indicating that the impulse response component of the fixed excitation codebook in the Nth subframe is “carried over” as a carried component error. While outputting to the evaluation mode determination part 71 (step ST53), the index k0 preserve | saved previously is output to the carry-over component error evaluation mode determination part 71 (step ST54).

When the signal C _previous is smaller than a predetermined threshold, the mode-specific minimum error search unit 67 sets the NSF mode information indicating that the impulse response component of the fixed excitation codebook in the Nth subframe is not carried over to the carryover component error. While outputting to the evaluation mode determination part 71 (step ST55), the index k1 preserve | saved previously is output to the carry-over component error evaluation mode determination part 71 (step ST56).

When the carry-over component error evaluation mode determination unit 71 receives the mode information of the NSF indicating “carry-over” from the mode-specific minimum error search unit 67, the carry-over component error evaluation mode determination unit 71 adds the impulse response component of the fixed excitation codebook from the buffer 70. Mode information (determination value) to be indicated and pulse information (determination value) corresponding to the value of k indicated by the index k0 are acquired, and the mode information (determination value) and pulse information (determination value) are transmitted to the multiplexing unit 72. Output.
Further, when mode information of the NSF indicating “no carryover” is received from the mode-specific minimum error search unit 67, mode information (decision value) indicating that the impulse response component of the fixed excitation codebook is not added from the buffer 70; The pulse information (determination value) corresponding to the value of k indicated by the index k1 is acquired, and the mode information (determination value) and pulse information (determination value) are output to the multiplexing unit 72.

  The multiplexing unit 72 multiplexes the mode information (determination value) and pulse information (determination value) output from the carry component error evaluation mode determination unit 71 and the multiplexed signal output from the buffer 69, and multiplexes the multiplexed information. The encrypted signal is transmitted to the speech decoding apparatus.

  According to the eleventh embodiment, whether to carry over to the subframe immediately after the subframe ((N + 1) th subframe) is left to the processing of the (N + 1) th subframe (delayed decision). As a result, there is no mismatch between whether or not the evaluation section can be extended and whether or not carryover is possible, and it is possible to determine the necessity / unnecessity of carryover after the evaluation of the carryover component, thereby improving the sound quality. It becomes possible.

Embodiment 12 FIG.
FIG. 29 is a block diagram showing a part of a speech encoding apparatus according to Embodiment 12 of the present invention. In the figure, the same reference numerals as those in FIG.
The gain storage memory 73 is a memory for storing the gain (gain for the excitation source signal of the fixed excitation codebook 10) output from the gain codebook 10.
The gain multiplier 74 multiplies the signal C _previous which is the impulse response component of the fixed excitation codebook 9 in the previous subframe stored in the carry component storage memory 23 by the gain stored in the gain storage memory 73. Perform the process.
The gain storage memory 73 and the gain multiplier 74 constitute addition means.

30 is a block diagram showing a part of a speech decoding apparatus according to Embodiment 12 of the present invention. In the figure, the same reference numerals as those in FIG.
The gain storage memory 85 is a memory for storing the gain (gain for the excitation excitation signal of the fixed excitation codebook 33) output from the gain codebook 34.
The gain multiplier 86 multiplies the signal C _previous which is the impulse response component of the fixed excitation codebook 33 in the previous subframe stored in the carry component storage memory 42 by the gain stored in the gain storage memory 85. Perform the process.
The gain storage memory 85 and the gain multiplier 86 constitute addition means.

Next, the operation will be described.
In the speech coding apparatus, as in the fourth embodiment, the carry component storage memory 23 stores the impulse response component of the fixed excitation codebook 9 in the previous subframe.
For example, in the case of the pulse A selected in the (N−1) th subframe (see FIG. 10), the impulse response component (section A = C _{previous in} FIG. 10) carried over to the Nth subframe is the (N -1) Since the pulse A is determined when the subframe is processed, the carry component storage memory 23 stores the signal C _previous .
The gain storage memory 73 stores the gain (gain for the excitation source signal of the fixed excitation codebook 10) output from the gain codebook 10 to the gain multiplier 12.

When the processing shifts from the (N−1) th subframe to the Nth subframe, the gain multiplier 74 extracts the signal C _previous from the carry component storage memory 23 and stores it in the gain storage memory 73 for the signal C _previous . The stored gain is multiplied and the multiplication result is output to the adder 24.

ただし、ｇ_currは、現フレームの固定音源符号帳９の利得（非量子化利得）である。 In the twelfth embodiment, when the minimum error search unit 20 searches for the value of k that maximizes the error evaluation R (k) of the equation (2), the error evaluation R is calculated using the following equation (12). The molecular component C (k) of (k) is calculated.

Here, g _curr is the gain (unquantized gain) of the fixed excitation codebook 9 of the current frame.

C _previous, which is the second term of C (k), can be obtained using Equation (11).
C _previous is a carry component of the fixed excitation codebook vector 9 of the (N−1) th subframe, and is a signal stored in the carry component storage memory 23.
Further, g _prev hat is a gain (quantization gain) of the fixed excitation codebook 9 of the previous frame, and is a signal stored in the gain storage memory 73.
Further, g _curr hat is a gain (quantization gain) of the fixed excitation codebook 9 of the current frame, and is a signal that is a codebook search target together with _ck .

In the speech decoding apparatus, in the processing of the (N−1) th subframe, pulse information indicating the position of the pulse A is transmitted from the speech encoding apparatus, but the fixed excitation codebook 33 is fixed to the speech encoding apparatus. Since it has the same filter as the excitation codebook 9 (impulse response information of the same internal filter as the fixed excitation codebook 9), an impulse response component that is carried forward to the Nth subframe can be automatically obtained. The storage memory 42 stores the signal C _previous which is an impulse response component.
At the same time, the gain storage memory 85 stores the gain for the excitation source signal of the fixed excitation codebook 33 decoded by the gain codebook 34.

When the processing shifts from the (N−1) th subframe to the Nth subframe, the gain multiplier 86 extracts the signal C _previous from the carry component storage memory 42 and stores it in the gain storage memory 85 for the signal C _previous . The stored gain is multiplied and the multiplication result is output to the adder 43.

  With the above configuration, the gain of the carry-over component accurately reflects the remaining impulse response component from the previous subframe, so the subframe immediately before the subframe ((N−1) th subframe). ), The effect of mismatch due to the gain of the current subframe being applied to the carry-over component from the previous subframe can be avoided, especially when the gain changes rapidly over the subframe. Improvements can be made.

ｇ_currは、現フレームの固定音源符号帳の利得（非量子化利得）である。 However, in the twelfth embodiment, the quantization of the fixed sound source (waveform) and the quantization of the gain must proceed simultaneously, and the search process may be enormous. For example, the following equation (13) ), Instead of the quantization gain of the current subframe, the fixed excitation codebook search is performed using the gain g _curr of the fixed _excitation component before quantization determined by the input signal as an approximate value. For example, although there is a slight deterioration in sound quality, the search processing can be reduced by separating the search for the fixed excitation codebook and the search for the gain codebook.

g _curr is the gain (non-quantization gain) of the fixed excitation codebook of the current frame.

It is a block diagram which shows the audio | voice coding apparatus by Embodiment 1 of this invention. It is a block diagram which shows the audio | voice decoding apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the waveform of an input audio | voice, the frame and sub-frame of an input audio | voice, a reference signal, etc. It is explanatory drawing which shows the area where evaluation of a quantization error is performed. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 2 of this invention. It is explanatory drawing which shows the difference in the weighting with respect to a quantization error. It is explanatory drawing which shows the waveform of an input audio | voice, the frame and sub-frame of an input audio | voice, a reference signal, etc. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 4 of this invention. It is a block diagram which shows the audio | voice decoding apparatus by Embodiment 4 of this invention. It is explanatory drawing which shows the impulse response component etc. carried over by the Nth sub-frame. It is a block diagram which shows a part of speech coding apparatus by Embodiment 5 of this invention. It is a block diagram which shows a part of speech decoding apparatus by Embodiment 5 of this invention. It is a block diagram which shows a part of speech coding apparatus by Embodiment 6 of this invention. It is a block diagram which shows a part of speech decoding apparatus by Embodiment 6 of this invention. It is explanatory drawing which shows the waveform of an input audio | voice, the frame and sub-frame of an input audio | voice, a reference signal, etc. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 7 of this invention. It is a block diagram which shows the audio | voice decoding apparatus by Embodiment 7 of this invention. It is a flowchart which shows the outline of the processing content of the audio | voice coding apparatus by Embodiment 7 of this invention. It is a block diagram which shows a part of speech coding apparatus by Embodiment 8 of this invention. It is a block diagram which shows a part of speech decoding apparatus by Embodiment 8 of this invention. It is a block diagram which shows the internal structural example of the pitch emphasis filter 63 of FIG. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 9 of this invention. It is a block diagram which shows the audio | voice decoding apparatus by Embodiment 9 of this invention. It is a block diagram which shows the internal structural example of the pitch emphasis filter 65 of FIG. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 10 of this invention. It is a flowchart which shows the outline of the processing content of the audio | voice coding apparatus by Embodiment 10 of this invention. It is a block diagram which shows the audio | voice coding apparatus by Embodiment 11 of this invention. It is a flowchart which shows the outline of the processing content of the audio | voice coding apparatus by Embodiment 11 of this invention. It is a block diagram which shows a part of speech coding apparatus by Embodiment 12 of this invention. It is a block diagram which shows a part of speech decoding apparatus by Embodiment 12 of this invention. It is a block diagram which shows the conventional audio | voice encoding apparatus. It is a block diagram which shows the conventional speech decoding apparatus. It is explanatory drawing which shows the waveform of an input audio | voice, and the frame and sub-frame of an input audio | voice. It is a block diagram which shows the fixed excitation codebook which a pulse excitation codebook uses. It is a block diagram which shows the fixed excitation codebook which uses a pulse-spreading codebook. It is a block diagram which shows the fixed excitation codebook which uses a pulse-spreading codebook. It is explanatory drawing which shows the wave packet shape which has predetermined | prescribed time length. It is explanatory drawing explaining the problem in the case of performing a CELP encoding process using the wave packet which has a predetermined | prescribed time length.

Explanation of symbols

  1 buffer, 2 preprocessing unit, 3 spectrum analysis unit (spectrum analysis unit), 4 linear prediction analysis unit, 5 LSP codebook, 6 LSP quantization / inverse quantization unit, 7 drive excitation generation unit (drive excitation generation unit) 8 adaptive codebook, 9 fixed excitation codebook, 10, 65c gain codebook, 11, 12 gain multiplier, 13, 63a, 65a adder, 14 synthesized speech generation unit (synthesized speech generation means), 15 LSP / LPC Conversion unit, 16 synthesis filter, 17 reference vector assembly buffer (parameter search unit), 18 subtractor (parameter search unit), 19 auditory weighting filter (parameter search unit), 20 minimum error search unit (parameter search unit), 21, 29, 72 Multiplexing unit, 22 Weighted minimum error searching unit (parameter searching means), 23, 42 Carrying forward component storage memo (Adding means), 24, 43 Adder (adding means), 25, 44, 74, 86 Gain multiplier (adding means), 26, 28, 45, 47 Switch (adding means), 27, 64, 66 Mode selection Minimum error search section with function (parameter search means), 31, 46 Demultiplexing section (information receiving means), 32 Adaptive codebook (drive excitation generator), 33 Fixed excitation codebook (drive excitation generator), 34 Gain code Book (driving excitation generating means), 35, 36 Gain multiplier (driving excitation generating means), 37 Adder (driving excitation generating means), 38 LSP codebook (synthesized speech decoding means), 39 LSP / LPC converter (synthesis) Speech decoding unit), 40 synthesis filter (synthetic speech decoding unit), 41 post filter (synthetic speech decoding unit), 61, 81 pitch stability evaluation unit (adding unit), 62, 63, 6 5, 82, 83, 84 Pitch emphasis filter (adding means), 63b, 65b Inverse filter, 63c, 65d Filter coefficient multiplier, 67 Mode-specific minimum error searching section (parameter searching means), 68 Multiplexing section (parameter searching means) ), 69, 70 buffer (parameter search means), 71 carry-over component error evaluation mode determination section (parameter search means), 73, 85 gain storage memory (addition means).

Claims (20)

  Drive that generates a plurality of drive sound sources by combining spectrum analysis means for performing sound spectrum analysis for each frame of input speech and excitation sound source signals in units of subframes output from the adaptive code book and fixed sound source code book A synthesis filter is formed according to a spectrum analysis result of the sound source generation means and the spectrum analysis means, and a plurality of driving sound sources generated by the driving sound source generation means are passed through the synthesis filter to generate a plurality of synthesized sounds. A synthesized speech generating means, a plurality of synthesized speech generated by the synthesized speech generating means and a quantization error of the input speech, and a code related to the synthesized speech having the smallest quantization error among the plurality of synthesized speech In the speech coding apparatus comprising the parameter search means for searching for the conversion parameter, the adaptive codebook and the fixed excitation codebook correspond to the subcode. Output the excitation source signal of the next subframe in addition to the excitation source signal of the next subframe, and the parameter search means uses not only the input speech of the subframe but also the input speech of the next subframe as an object of quantization error A speech encoding apparatus comprising:   The parameter search means sets the weighting for the quantization error related to the input speech of the subframe larger than the weighting for the quantization error related to the input speech of the next subframe, and encodes the synthesized speech with the minimum quantization error. The speech coding apparatus according to claim 1, wherein a parameter is searched.   2. The speech coding according to claim 1, wherein the parameter search means excludes the input speech of the next subframe from the quantization error evaluation target when the next subframe belongs to a frame different from the subframe. apparatus.   For each frame of the input speech, spectrum analysis means for performing speech spectrum analysis, the excitation source signal of the subframe output from the fixed excitation codebook and the response component of the fixed excitation codebook in the previous subframe are added Drive excitation source that generates a plurality of drive excitation sources by combining the excitation excitation signal of the fixed excitation codebook to which the response component is added by the addition means and the excitation excitation signal of the subframe output from the adaptive codebook A synthesis unit that forms a synthesis filter according to a spectrum analysis result of the generation unit and the spectrum analysis unit, and passes a plurality of driving sound sources generated by the driving sound source generation unit through the synthesis filter to generate a plurality of synthesized speech Evaluating quantization errors of speech generation means, a plurality of synthesized speech generated by the synthesized speech generation means, and the input speech Speech encoding apparatus that includes a parameter search means quantization error among the plurality of synthesized speech is to search for the coding parameters of the smallest synthetic speech.   The adding means multiplies the response component of the fixed excitation codebook in the previous subframe by a gain of 0 or more and less than 1, and then adds the response component to the excitation excitation signal of the subframe. Speech encoding device.   When the previous subframe belongs to a frame different from the subframe, the adding means adds the response component of the fixed excitation codebook in the previous subframe to the excitation excitation signal of the subframe output from the fixed excitation codebook. 5. The speech coding apparatus according to claim 4, wherein no addition is performed.   The parameter search means determines whether or not to add the response component of the fixed excitation codebook in the previous subframe based on the quantization error evaluation result, and indicates that the determination result of the parameter search means is added. 5. The speech coding apparatus according to claim 4, wherein the adding means adds the response component to the excitation sound source signal of the subframe.   When a pitch emphasis filter is added to the fixed excitation codebook, the adding means filters the pitch enhancement filter for the response component of the fixed excitation codebook in the previous subframe according to the stability of the pitch period of the past subframe. 5. The speech coding apparatus according to claim 4, wherein the effect is adjusted.   When the pitch enhancement filter is added to the fixed excitation codebook, the parameter search means adjusts the filter effect of the pitch enhancement filter on the response component of the fixed excitation codebook in the previous subframe, and minimizes the quantization error. 5. The speech coding apparatus according to claim 4, wherein coding parameters relating to speech are searched.   The adding means excites the subframe output from the fixed excitation codebook only when the parameter search means includes not only the input speech of the subframe but also the input speech of the next subframe as an object of quantization error evaluation. 5. The speech coding apparatus according to claim 4, wherein the response component of the fixed excitation codebook in the previous subframe is added to the excitation signal.   The parameter search means excludes the input speech of the next subframe from the evaluation target of the quantization error when the minimum quantization error when the input speech of the next subframe is included in the evaluation target is larger than a predetermined value. The speech encoding apparatus according to claim 10.   11. The parameter search means, based on a quantization error evaluation result in the next subframe, determines whether or not to include the input speech of the next subframe as a quantization error evaluation target. Speech encoding device.   5. The speech encoding apparatus according to claim 4, wherein the adding means stores a gain of the fixed excitation codebook with respect to the excitation excitation signal, and multiplies the gain by the response component of the fixed excitation codebook in the previous subframe.   Spectral information indicating the spectral analysis result of speech from the speech coding device, pitch information of the excitation source signal output from the adaptive codebook when a synthesized speech with minimum quantization error is obtained, and synthesis with minimum quantization error Information receiving means for receiving pulse information of the excitation excitation signal output from the fixed excitation codebook when speech is obtained, and the fixed excitation codebook in the subframe corresponding to the pulse information received by the information receiving means An addition means for adding the excitation source signal and the response component of the fixed excitation codebook in the previous subframe, an excitation excitation signal of the fixed excitation codebook to which the response component is added by the addition means, and the information receiving means Driving excitation generating means for generating a driving excitation from the excitation excitation signal of the adaptive codebook in the subframe corresponding to the pitch information; Synthetic speech decoding that forms a synthesis filter according to the spectrum analysis result indicated by the spectrum information received by the information receiving means, passes the driving sound source generated by the driving sound source generating means through the synthesis filter, and decodes the synthesized speech And a speech decoding apparatus.   The adding means multiplies the response component of the fixed excitation codebook in the previous subframe by a gain of 0 to less than 1, and then adds the response component to the excitation signal of the fixed excitation codebook in the subframe. The speech decoding apparatus according to claim 14.   When the previous subframe belongs to a frame different from the subframe, the adding means does not add the response component of the fixed excitation codebook in the previous subframe to the excitation signal of the fixed excitation codebook in the subframe. The speech decoding apparatus according to claim 14.   The information receiving means receives mode information indicating whether or not to add the response component of the fixed excitation codebook in the previous subframe from the speech encoding apparatus, and the adding means adds the mode information received by the information receiving means. 15. The speech decoding apparatus according to claim 14, wherein the response component is added to the excitation signal of the fixed excitation codebook in the subframe.   When a pitch emphasis filter is added to the fixed excitation codebook, the adding means filters the pitch enhancement filter for the response component of the fixed excitation codebook in the previous subframe according to the stability of the pitch period of the past subframe. The speech decoding apparatus according to claim 14, wherein an effect is adjusted.   When the pitch enhancement filter is added to the fixed excitation codebook, the information receiving means receives mode information indicating the state of the pitch enhancement filter when the synthesized speech with the minimum quantization error is obtained from the speech encoding device, 15. The speech decoding apparatus according to claim 14, wherein the adding means adjusts the filter effect of the pitch enhancement filter for the response component of the fixed excitation codebook in the previous subframe according to the mode information received by the information receiving means.   15. The speech decoding apparatus according to claim 14, wherein the adding means stores a gain of the fixed excitation codebook for the excitation excitation signal, and multiplies the gain by the response component of the fixed excitation codebook in the previous subframe.

Images