JPH09297598A - Voice coding/decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make the same sound source vector obtainable on both the coder side and the decoder side even immediately after returning from a transmission line error and to mitigate the error of the sound source vector occurring on both the coder side and the decoder side occurring immediately after the trans mission line error. SOLUTION: A searching scope limiter 109 judges whether a transmission line error is generated in the just previous frame or subframe from a transmission line error monitoring code 110 inputted from the decoder side. When the error is judged to be generated in the just previous frame or subframe, the portion which is generated in the just previous frame in the sound source signals generated in the past and stored in an adaption code book 113 is excluded from the searching scope and subjected to searching which uses an adaption code book to output the searching scope of the book to an error minimizing means 108 so as to select the optimum code vector. When the transmission line error continues, the vector gain of the adation code book 113 is nullified or switched to another fixed code book.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a CELP type voice encoding / decoding device.

[0002]

2. Description of the Related Art In recent years, the demand for digital mobile communications has increased, and it has become necessary to reduce the bit rate of voice encoding, and various voice encoding devices have been developed. Among them, in the CELP method, a voice signal is separated into vocal tract information and sound source information, the vocal tract information is expressed by a digital filter composed of linear prediction coefficients, and the sound source information is made into a waveform pattern of several hundred to several thousand types. Vector quantization using an excitation codebook composed of
/ S to 8 kb / s) is widely used as a method capable of realizing high quality voice.

A CELP system excitation is composed of excitation vectors selected from two types of codebooks, an adaptive codebook and a fixed codebook (probabilistic codebook and random codebook). this house,
The adaptive codebook expresses the periodic component contained in the excitation signal (particularly the vowel part), and stores the excitation signal waveform synthesized in the past. On the other hand, the fixed codebook is facilitated in advance to represent a random waveform (random component of the excitation signal) after removing the periodic component from the excitation signal. The fixed codebook is created by random numbers, or is created by learning using a large number of voice data,
Many types, such as those composed of pulse trains, have been proposed and used. However, in a CELP type speech encoding device, since the excitation signal generated in the past is used as an adaptive codebook, a transmission path error is generated. When the error occurs, even after recovering from the error, since the excitation signal generated at the time of the error is stored as the adaptive codebook, there is a problem that the influence of the error propagates.

The adaptive codebook search unit in the conventional speech coding apparatus based on the CELP method will be described below.
FIG. 5 shows a general CELP type speech coder. In FIG. 5, the input voice signal 1 is the preprocessor 2
The waveform is shaped by and then output to the linear prediction analyzer 3 and the adder 4. The linear prediction analyzer 3 performs a linear prediction analysis using the preprocessed input speech signal and outputs a linear prediction coefficient to the synthesis filter 5. The synthesis filter 5 performs speech synthesis using the sound source signal input from the adder 6 and the linear prediction coefficient input from the linear prediction analyzer 3, and outputs the speech to the adder 4. The adder 4 calculates an error between the combined signal input from the combining filter and the preprocessed input audio signal input from the preprocessor 2, and outputs the error to the perceptual weighting unit 7.
The perceptual weighter 7 perceptually weights the error signal,
It outputs to the error minimizing means 8. The error minimization means 8 determines a fixed code vector, an adaptive code vector, a fixed code vector gain, and an adaptive code vector gain so that the auditory weighting error input from the auditory weighting device 7 is minimized. The fixed code vector is selected from the fixed code book 9 and output to the fixed code vector gain multiplier 10. The fixed code vector gain multiplier 10 multiplies the fixed code vector output from the fixed code book 9 by the fixed code vector gain, and outputs the product to the adder 6. The adaptive code vector is selected from the adaptive codebook 11 and output to the adaptive code vector gain multiplier 12. Adaptive code vector gain multiplier 1
2 multiplies the adaptive code vector output from the adaptive codebook 11 by the adaptive code vector gain, and outputs the product to the adder 6. The adder 6 adds the respective vectors output from the fixed code vector gain multiplier 10 and the adaptive code vector gain multiplier 12, and outputs the addition to the synthesis filter 5 as the excitation vector. Fixed code vector with minimum error,
The excitation vector generated by the adder 6 by the combination of the adaptive code vector, the fixed code vector gain, and the adaptive code vector gain is newly added to the adaptive codebook buffering the past excitation signal. Then, information on the adaptive code vector, the fixed code vector, the adaptive code vector gain, and the fixed code vector gain that generate the excitation vector that minimizes this error is transmitted to the decoder side.

[0005]

However, in the above-mentioned conventional CELP-type speech coding apparatus, when a transmission path error occurs, the contents buffered in the adaptive codebook are displayed on the encoder side and the decoder side. There is a problem in that they are different and are greatly affected by the error even after recovering from the transmission line error.

The present invention solves the above-mentioned conventional problems, and enables an encoder side and a decoder side to obtain the same excitation vector immediately after recovering from a transmission path error, and recovering from a transmission path error. An object of the present invention is to provide a speech encoding / decoding device capable of mitigating an error between excitation vectors generated on the encoder side and the decoder side immediately after the operation.

[0007]

According to the present invention, in order to achieve the above object, information for monitoring occurrence of a transmission line error is sent from a decoder side to an encoder side, and a transmission line error occurs. If it is determined that the search range of the adaptive codebook search in the next frame or subframe in which an error has occurred is limited. Further, when transmission line errors occur continuously, the coding process using only the fixed codebook is continued until the transmission line error is eliminated without using the adaptive book. Furthermore, the use of the adaptive codebook generated when a transmission line error occurs is avoided, and the adaptive codebook generated when there is no transmission line error is used.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION According to the first aspect of the present invention, an encoder which receives transmission path error information from a decoder side judges whether a transmission path error has occurred in the immediately preceding frame or subframe. When a transmission path error occurs, the adaptive codebook search range limiting means for excluding the portion generated in the immediately preceding frame or subframe from the search range of the adaptive codebook is provided. It is possible to generate the same excitation vector on the encoder side and the decoder side also in the frame or subframe.

According to the second aspect of the present invention, when it is determined that the transmission path error has occurred over the immediately preceding several frames or several subframes, the frames or the frames in which the transmission path error has occurred continuously or The gain of the adaptive code vector generated in the subframe is set to zero, and the excitation vector is generated only from the fixed codebook.Even in the frame or subframe immediately after the transmission path error is resolved, the encoder side and the decoder side have It is possible to generate the same sound source vector.

According to the third aspect of the present invention, when it is determined that the transmission path error has occurred over the immediately preceding several frames or several subframes, the frames in which the transmission path error has occurred successively. Alternatively, the adaptive code codebook generated in the subframe is switched to the fixed codebook, and means for generating the excitation vector from only the fixed codebook is provided. It is possible to generate the same excitation vector on the device side and the decoder side.

According to a fourth aspect of the present invention, when a decoder receives normal information from the encoder side according to the second or third aspect after a transmission line error has occurred, the adaptive codebook Instead, the fixed codebook is used for speech synthesis, and the same excitation vector can be generated on the encoder side and the decoder side even in the frame or subframe immediately after the transmission path error is resolved.

According to a fifth aspect of the present invention, when normal information is received from the encoder side after a transmission path error has occurred, the decoder determines that the received pitch sound source is an error frame or subframe. It is checked whether the synthesized excitation sound source information is included, and if it is not included, the pitch information is generated from the adaptive codebook using the received information as it is, and if it is included, the pitch sound source is used without using the received information as it is. Since it is generated, it is possible to avoid a large error between excitation vectors obtained on the encoder side and the decoder side.

(Embodiment 1) Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a CELP type speech coding apparatus according to the first embodiment of the present invention. In FIG. 1, 101 is an input speech signal, 102 is an input speech signal 101, and the input speech signal after pre-processing is a linear predictor 103 and an adder 1.
104 is a preprocessor for outputting to 04, 103 is a linear prediction analyzer for inputting the input speech signal after preprocessing, and outputting a linear prediction coefficient to the synthesis filter 105;
Receives the pre-processed voice signal and the output signal of the synthesis filter 105 as input, calculates a difference signal, and the auditory weighting device 10
7 is an adder, 105 is a synthesizing filter for synthesizing a speech signal with the excitation vector output from the adder 106 and the linear prediction coefficient output from the linear prediction analyzer 103 as inputs, and 106 is a fixed code vector gain Multiplier 1
12 and an adder that outputs the respective vectors output from the adaptive code vector gain multiplier 114 and outputs the result to the synthesis filter 105, 107 performs auditory weighting using the error signal output from the adder 104 as an input, The perceptual weighting unit that outputs the error to the error minimization unit 108 includes a combination of a fixed code vector, an adaptive code vector gain, and an adaptive code vector gain that minimizes the perceptual weighting error power output from the perceptual weighting unit 107.
Error minimization means for deciding on the basis of the search range output from the search range limiter 109, 109 receives the transmission path error monitoring signal as an input, and the error minimization means 108 decides the search range for adaptive codes and the like to minimize the error. The search range limiter output to the conversion means 108, 110 a transmission path error monitoring signal for detecting the occurrence of a transmission path error, and 111 a fixed number of fixed code vectors output to the fixed code vector gain multiplier 112. , 112 is a fixed code vector gain multiplier that outputs the fixed code vector output from the fixed code book 111 by a fixed vector gain to the adder 106, and 113 is the past output from the adder 106. Of the sound source vector (finally determined by the error minimization means 108) of Adaptive codebook for outputting the adaptive code vector gain multiplier 114 as an adaptive code vector by cutting out part, 114 denotes an adder multiplied by the adaptive code vector gain to the adaptive code vector outputted from adaptive codebook 113 1
It is an adaptive code vector gain multiplier for outputting to 06.

The operation of the CELP type speech coding apparatus configured as described above will be described below. FIG.
In, the input speech signal 101 is a digital signal consisting of a predetermined number of samples, and the speech coding process is
This is performed for each audio signal of the predetermined number of samples.
The audio signal block having the predetermined number of samples is called a frame or subframe. The input audio signal 101 is
Bandwidth limitation and gain adjustment are performed by the preprocessor 102. The linear prediction analyzer 103 performs publicly known linear prediction analysis using the audio signal after the pre-processing, and calculates a linear prediction coefficient. The synthesis filter 105 includes the linear prediction analyzer 10
A filter is formed by using the linear prediction coefficient calculated in step 3, and the sound source vector output from the adder 6 is filtered to give a voice. The adder 104 calculates a difference signal between the preprocessed input audio signal and the audio signal synthesized by the synthesis filter 105. The perceptual weighter 107 perceptually weights the difference signal calculated by the adder 104, and outputs the weighted difference signal to the error minimizing means 108. This perceptual weighting is generally performed using a filter in which a linear prediction filter calculated by the linear prediction analyzer 103 and a linear prediction filter using the perceptual weighting coefficient are connected in series. The error minimization means 108 converts the excitation vector input to the synthesis filter 105 into a fixed code vector, a fixed code vector gain, and an adaptive code vector so that the power of the difference signal (error signal) after the auditory weighting is minimized. Adjust by changing the combination of adaptive code vector gains. Generally, first, the optimum adaptive code vector is extracted from the adaptive codebook 113,
The multiplier 114 multiplies the adaptive code vector gain to determine the output to the adder 106, and then extracts the optimum fixed code vector when combined with the adaptive code vector from the fixed code book 111, and the multiplier At 112, the fixed code vector gain is multiplied to determine the output to the adder 106. The search range limiter 109 limits the search range of the adaptive codebook 113 when extracting the optimum adaptive code vector from the adaptive codebook 113. The search range limiter 109 determines from the transmission line error monitor signal 110 input to the search range limiter 109 whether a transmission line error has occurred in the immediately preceding frame or subframe. And
When it is determined that the transmission path error has occurred in the immediately preceding frame or subframe, the part generated in the immediately preceding frame among the excitation signals generated in the past stored in the adaptive codebook 113 is removed from the search range. The adaptive codebook search is performed, and the search range of the adaptive codebook 113 is output to the error minimizing means 108 so as to select the optimum code vector.
If it is determined that a transmission path error has occurred consecutively in the immediately preceding frame or subframe, the adaptive codebook 113
The adaptive codebook search range is determined so as to perform the adaptive codebook search by removing the part generated in the immediately preceding continuous frame from the excitation signal stored in the past stored in the To 108. However, if all of the excitation codebooks stored in adaptive codebook 113 are excluded from the search range due to a long series of transmission path errors, the adaptive code vector gain is set to zero. The search range of the error minimizing means 108 is determined so that the excitation vector is generated only from the fixed code vector.

When the speech coding apparatus is configured as described above, it is necessary to add means for transmitting the transmission path error monitoring signal 110 to the decoding apparatus, but the decoding processing in the coding apparatus is required. Is the same as the conventional one, the conventional one can be used as it is. Note that the transmission path error monitoring signal 110 may be a signal that is transmitted at a predetermined time interval (shorter than the time interval for transmitting the coding parameter for one frame), or the like. In this case, the search range is When the limiter 209 receives a signal different from the predetermined signal, it determines that a transmission path error has occurred in the encoded information of the frame transmitted at that time.

As described above, according to the first embodiment, the encoder that receives the transmission path error information from the decoder side determines whether or not a transmission path error has occurred in the immediately preceding frame or subframe, When a transmission path error occurs, a search range limiter 109 that excludes the portion generated in the immediately preceding frame or subframe from the search range of the adaptive codebook is provided. Also in the subframe, it is possible to generate the same excitation vector on the encoder side and the decoder side.

(Second Embodiment) Next, a second embodiment of the present invention will be described with reference to FIG. In FIG. 2, 201 is an input speech signal, 202 is a preprocessor for inputting the input speech signal 201, and outputting the preprocessed input speech signal to a linear prediction analyzer 203 and an adder 204, 20
Reference numeral 3 denotes a linear prediction analyzer that performs linear prediction analysis using the preprocessed input speech signal as an input and outputs linear prediction coefficients to the synthesis filter 205. Reference numeral 204 denotes the preprocessed speech signal and the output signal of the synthesis filter 205. An adder that calculates a difference signal as an input and outputs the difference signal to the perceptual weighting unit 207, and 205 synthesizes a speech signal with the sound source vector output from the adder 206 and the linear prediction coefficient output from the linear prediction analyzer 203 as inputs. 207 is an output from the adder 204. 207 is an adder that adds the vectors output from the fixed code vector gain multiplier 212 and the adaptive code vector gain multiplier 216 and outputs the added vector to the synthesis filter 205. The perceptual weighting device 20 which receives the error signal as an input, performs perceptual weighting, and outputs it to the error minimizing means 208, Is a combination of a fixed code vector, an adaptive code vector, a fixed code vector gain, and an adaptive code vector gain output from the perceptual weighting unit 207 that minimizes the error power after the perceptual weighting is output from the search range limiter 209. Error minimization means 209 for determining the transmission path error monitoring signal 210
As an input, the adaptive codebook 2 by the error minimizing means 208
A search range limiter 210 which determines the search ranges of 14 and outputs them to the error minimization means 208 and the codebook selector 215.
Is a transmission line error monitoring signal for detecting the occurrence of a transmission line error, 211 is a fixed codebook for storing a fixed number of fixed code vectors for outputting a fixed code vector to the fixed code vector gain multiplier 212, 212 Is the fixed codebook 21
The fixed code vector gain multiplier 213 that multiplies the fixed code vector output from 1 by the fixed code vector gain and outputs the fixed code vector to the adder 206. The fixed code vector gain multiplier 213 outputs the fixed code vector to the codebook selector 215. A fixed codebook for storing code vectors, and 214 is a buffer of past excitation vectors (finally determined by the error minimization means 208) output from the adder 206, and stores the signal sequence stored in the buffer. The adaptive codebook 215, which cuts out a part and outputs it as an adaptive code vector to the codebook selector 215, inputs search range information from the search range limiter 209 and inputs vectors from the fixed codebook 213 and the adaptive codebook 214, respectively. A codebook selector that selects only one of them and outputs it to the code vector gain multiplier 216;
A code vector gain multiplier 6 multiplies the code vector output from the code book selector 215 by a code vector gain and outputs the result to the adder 206.

The operation of the CELP type speech coder constructed as above will be described below. In FIG. 2, the input voice signal 201 is a digital signal having a predetermined number of samples, and the voice encoding process is performed for each voice signal having the predetermined number of samples. The audio signal block having the predetermined number of samples is called a frame or subframe. The input audio signal 201 is band-limited and gain-adjusted by the preprocessor 202.
Using the speech signal after this pre-processing, the linear prediction analyzer 20
3 performs a known linear prediction analysis to calculate a linear prediction coefficient. The synthesis filter 205 forms a filter by using the linear prediction coefficient calculated by the linear prediction analyzer 203, and filters the sound source vector output from the adder 206 to perform speech synthesis. The adder 204 calculates a differential signal between the preprocessed input audio signal and the audio signal synthesized by the synthesis filter 205. The perceptual weighting unit 207 perceptually weights the difference signal calculated by the adder 204, and the error minimizing means 208 is provided.
Output to This perceptual weighting is generally performed using a cascaded filter of linear prediction filters using the linear prediction coefficient calculated by the linear prediction analyzer 203 and the perceptual weighting coefficient. The error minimizing means 208 is
The excitation vector input to the synthesis filter 205 is a combination of a fixed code vector, a fixed code vector gain, an adaptive code vector, and an adaptive code vector gain so that the power of the difference signal (error signal) after perceptual weighting is minimized. Adjust by changing. In general, first, the optimum adaptive code vector is extracted from the adaptive code book 214, multiplied by the adaptive code vector gain in the multiplier 216 to determine the output to the adder 206, and then the fixed code book 21.
A fixed code vector that is optimal when combined with an adaptive vector is taken out of 1 and multiplied by the fixed code vector gain in the multiplier 212 to determine the output to the adder 206. The search range limiter 209 limits the search range of the adaptive codebook 214 when extracting the optimum adaptive code vector from the adaptive codebook 214. The search range limiter 209 determines from the transmission line error monitor signal 210 input to the search range limiter 209 whether a transmission line error has occurred in the immediately preceding frame or subframe. Then, when it is determined that a transmission path error has occurred in the immediately preceding frame or subframe, of the excitation signals generated in the past stored in adaptive codebook 214, the portion generated in the immediately preceding frame is extracted from the search range. The adaptive codebook search is performed by removing the search range, and the search range of the adaptive codebook 214 is output to the error minimizing means 208 and the codebook selector 215 so that the optimum code vector is selected. When it is determined that a transmission path error has occurred in the immediately preceding frame or subframe in succession, the part generated in the immediately preceding continuous frame in the excitation signal generated in the past stored in the adaptive codebook 214 is detected. The adaptive codebook search range is determined so as to perform the adaptive codebook search outside the search range, and output to the error minimizing means 208. However, when all the excitation codebooks stored in the adaptive codebook 214 are excluded from the search range due to the continuous transmission error for a long time,
The search range is output to the codebook selector 215 and the error minimizing means 208 so that the fixed codebook 213 is used instead of the adaptive codebook 214 to generate the excitation vector. When the input search range has a content indicating a fixed codebook search, the codebook selector 215 calculates the code vector input from the fixed codebook 213 by using the code vector gain multiplier 2
Output to 16.

When the speech coding apparatus is configured as described above, the decoding apparatus needs a means for selecting either the adaptive codebook or the fixed codebook. As a simple method, information indicating which codebook is used may be added on the encoding device side and transmitted to the decoding device side. For this reason, if it is impossible to allocate bits, a transmission line error monitoring means is added to switch between the fixed codebook and the adaptive codebook when continuous transmission line errors have occurred in the past. I need to do it.

As the transmission path error monitoring signal 210, it is conceivable that a predetermined signal is transmitted at a constant time interval (shorter than the time interval for transmitting the coding parameter for one frame). In this case, When a signal different from a predetermined signal is received, search range limiter 209 determines that a transmission path error has occurred in the coded information of the frame transmitted at that time.

As described above, according to the second embodiment, in the normal frame after the error frame, the excitation vector of the encoding device and the excitation vector of the decoding device are the same without distortion. Can be obtained. Further, the adaptive codebook can be canceled to improve the sound quality as compared with the first embodiment in which the information amount assigned to the adaptive code vector is not used.

(Embodiment 3) Next, a speech decoding apparatus according to a third embodiment of the present invention will be described. FIG. 3 shows excitation waveforms stored in the adaptive codebook of the speech decoding apparatus, 301 is an excitation waveform stored in the adaptive codebook, and 302 is a frame that was not correctly decoded due to a transmission path error. , Pi is the lag value of the normal frame transmitted from the encoder immediately after the frame with the transmission path error, and Vpi is the adaptation cut out from the adaptive codebook based on the lag value Pi. The code vector section, NVpi, indicates the section (current frame or subframe) in which the excitation waveform is to be generated.

In FIG. 3, the adaptive code vector (interval V) represented by the lag value Pi transmitted from the encoder is shown.
Since pi) includes the excitation vector of the frame that was not correctly decoded due to the transmission path error, the waveform distortion becomes large. Therefore, if there is a transmission path error in the immediately preceding frame, an adaptive code vector is generated using a pitch (nPi) that is an integral multiple of the lag value Pi transmitted from the encoding device. The integer n at this time is the minimum value of the integers necessary for not including the sound source portion 302 generated by incorrect information, and in FIG. 3, n =
3, and Vp3 is cut out as an adaptive code vector.

FIG. 4 shows the excitation waveform of the adaptive codebook of the decoding device in the frame or subframe immediately after FIG. At this time, the adaptive code vector Vpi + 1 cut out based on the lag value Pi + 1 transmitted from the coding apparatus still includes the excitation waveform generated in the error frame. To avoid this, nPi +
Vp4 may be used as the adaptive code vector when n = 4 of 1. However, in the case shown in FIG. 4, Vp
Since the ratio of the source waveform generated in the error frame included in i + 1 is low, Vpi + 1 is used without using Vp4.
May be used, but in that case, it is necessary to perform case classification based on the ratio including the sound source waveform generated in the error frame included in Vpi + 1.

It is to be noted that such a method using an integer multiple pitch is effective in a voiced part where the pitch period is clear, and the adaptive code vector gain transmitted from the encoder is close to 1.0. Or the change in the lag value in several frames before the error frame is small and equal to or almost equal to the lag value in the normal frame immediately after the error frame. Further, since the adaptive code vector is cut out from the adaptive codebook while avoiding the part generated when an error has occurred in the past, the excitation waveform stored in the adaptive codebook of the decoding device has a longer time than the adaptive codebook of the coding device. Must be stored.

As described above, according to the third embodiment, the distortion of the adaptive code vector in the normal frame after the error frame can be suppressed in the portion where the adaptive code vector works effectively.

[0027]

As described above, according to the present invention, in the CELP type speech encoding / decoding device, the same excitation vector can be obtained on the encoder side and the decoder side even immediately after recovering from the transmission path error. Further, it is possible to realize an excellent speech coding / decoding device capable of mitigating the error between the excitation vector generated on the encoder side and the error generated on the decoder side immediately after the recovery from the transmission path error.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing a configuration of a speech encoding apparatus according to a second embodiment of the present invention.

FIG. 3 is a schematic diagram of an adaptive codebook of a speech encoding apparatus according to a third embodiment of the present invention.

FIG. 4 is a schematic diagram of an adaptive codebook of a speech encoding apparatus according to a third embodiment of the present invention.

FIG. 5 is a block diagram showing the configuration of a general CELP speech coding apparatus.

[Explanation of symbols]

104 adder 106 adder 112 fixed code vector gain multiplier 114 adaptive code vector gain multiplier 204 adder 206 adder 212 fixed code vector gain multiplier 216 code vector gain multiplier 301 adaptive code equal excitation waveform 302 error in frame Generated adaptive codebook excitation waveform section

Claims (5)

[Claims] 1. An encoder that receives transmission path error information from a decoder example determines whether a transmission path error has occurred in the immediately preceding frame or subframe, and if a transmission path error has occurred, the immediately preceding frame is detected. Alternatively, a speech coding / decoding device including adaptive codebook search range limiting means for excluding a portion generated in a subframe from the search range of the adaptive codebook. 2. When it is determined that a transmission path error has occurred over the immediately preceding several frames or several subframes, the gain of the adaptive code vector generated in the frame or subframe in which the transmission path error has occurred continuously. 2. The speech coding / decoding apparatus according to claim 1, wherein the excitation vector is generated only from the fixed codebook by setting 0 to zero. 3. When it is determined that a transmission path error has occurred over the immediately preceding several frames or several subframes, the adaptive codebook generated by the frames or subframes in which the transmission path error has occurred continuously. 2. The speech coding / decoding apparatus according to claim 1, further comprising means for switching from the fixed codebook to a fixed codebook and generating an excitation vector from the fixed codebook only. 4. When a decoder receives normal information from a coder according to claim 2 or 3 after a channel error has occurred, a fixed codebook is used instead of an adaptive codebook for speech synthesis. A voice encoding / decoding device for performing. 5. When normal information is received from the encoder side after a transmission path error has occurred, the decoder determines whether the received pitch excitation includes excitation excitation information combined in an error frame or subframe. A speech coding / decoding device that checks whether or not, and if not included, generates a pitch excitation from the adaptive codebook using the received information as it is, and if it does, generates a pitch excitation without using the received information as it is.