JPH0715353A - Voice decoder - Google Patents

Abstract

PURPOSE:To suppress deterioration in sound quality due to the error of non- protected code while the configuration is simplified. CONSTITUTION:A voice encoder 301 compresses a voice signal, classifies the voice codes depending on the importance and inputs them. An error correction encoder 302 protects the voice code with high importance through error correction. A multiplexer 303 multiplexes the protected code and a code not protected directly from the voice coder 301 to send the result to a channel 304. A multiplexer demultiplexer 305 demultiplexes the received and multiplexed code. An error correction decoder 306 decodes the protected code to obtain number of bit errors in the received code. An error rate calculation device 307 calculates an estimate error rate of the code not protected based on number of bit errors and an error protection processing unit 308 gives a command to a voice decoder 309 based on the estimate error rate to apply correction for voice decoding processing thereby suppressing deterioration in sound quality.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used in the field of voice signal communication and storage, and relates to a voice decoding apparatus for reducing deterioration of voice quality due to a code error generated during communication and storage. .

[0002]

2. Description of the Related Art In the field of communicating a voice signal and storing the voice signal in a semiconductor memory or a magnetic recording medium, a voice coding technique for digitizing the voice signal and compressing or transmitting the information for communication or storage, and communication or storage There is known a system combined with an error correction technique for reducing the deterioration of the voice quality due to a code error generated in the case of.

As a first conventional technique, a digital car telephone system in Japan will be described. Figure 3 shows the document "Channel Coding for Digital Speech Transmissio".
n in theJapanese Digital Cellular System ", MJMcLa
uhlin, The Institute of Electronics, Information and Communication Engineers, Research Group Material RCS90-27].

On the encoding side, 134 bits of the voice code output from the voice encoder 11 is 2 depending on its importance.
It is divided into one class (75bit and 59bit). Since the class 1 has a high degree of importance and the sound quality deteriorates if the code is erroneous, the CRC calculator 12 performs error detection coding, and then the convolutional encoder 13 performs error correction coding processing. On the other hand, the less important class 2 code is not protected by error correction and is transmitted as it is through the interleaver (multiplexer) 14.

If error detection / correction coding processing is performed on all speech-coded codes, the number of redundant bits for error correction becomes too large and the amount of information becomes large. Therefore, as in the above system, it is general to divide into a part to be subjected to the error detection / correction coding process and a part not to be subjected to the error detection / correction coding process according to the importance of the code. In the case of the digital car telephone system in Japan, about 44% (= 59/134) of less important information in the voice code is transmitted without protection.

Since no redundancy for error correction is added to the code of low importance, even if the code of high importance can be completely error-corrected by the correction process,
An error may occur in the less important code, and since it is not corrected, the sound quality is necessarily deteriorated.

On the decoding side, the processing opposite to that on the encoding side is performed for error correction. The processing procedure is as follows. After deinterleaving (separating) the received code, class 1 code is subjected to error correction decoding processing such as Viterbi decoding, and a CRC check (Cyclic Redundancy Check) is performed on the decoding result.
)I do. If the CRC check is correct, it means that an error-free code can be decoded for class 1. But if the CRC check is wrong,
Error protection processing such as squelch is performed to reduce the power so that noise is not noticeable in the reproduced audio signal.

As described above, according to the first prior art, the error correction coding process is performed under a certain error correction capability only for a portion of the voice code having high importance, and the error correction capability is within the error correction capability. Error is completely corrected, and if the error exceeds the error correction capability, appropriate error protection processing is performed.

As a second conventional technique, there is one that changes the error correction process and the voice coding process in accordance with the error state of the communication path. For example, there are JP-A-3-98318 (“Voice coding method”) and JP-A-62-117423 (“Voice coding method”).

In the case of the voice coding method disclosed in Japanese Patent Laid-Open No. 3-98318, on the encoder side, based on the current error rate information of the transmission path sent from the receiving side,
By changing the bit allocation of voice coding / error correction coding, processing according to the error state of the communication path is performed.

FIG. 4 shows a voice encoding system disclosed in Japanese Patent Laid-Open No. 62-117423. For convenience of explanation, the left side of the figure is A,
The right side is called B. 101 and 201 are information source coding units, 102 and 202 are error correction coding circuits, and 10
3, 203 is a framing circuit, 104, 204 are transmitters, 105, 205 are transmission paths (communication paths), 106, 20
6 is a receiver, 107 and 207 are deframing circuits, 1
08 and 208 are error correction decoding circuits, 109 and 209 are information source decoding units, 110 and 210 are error rate monitoring circuits, and 1
Reference numerals 11 and 211 denote control circuits.

The speech signal input to the information source coding unit 101 on the A side is coded and the error correction coding circuit 10 is used.
2 outputs the parameter information and the framing circuit 103 outputs the code corresponding to the prediction residual. After the parameter information is error-corrected by the error-correction coding circuit 102, it is added to the framing circuit 103 together with the prediction residual and converted into a serial code sequence. This serial code sequence is digitally converted by the transmitter 104, and then the wireless transmission line 10
5 is sent.

The information transmitted from the transmitter 104 on the A side is received by the receiver 106 on the B side via the transmission path 105. The deframing circuit 107 decomposes the input demodulation code sequence and outputs a code corresponding to the parameter information and the prediction residual. Of these, the parameter information is error-corrected and decoded in the error-correction decoding circuit 108, and then input to the information source decoding unit 109 together with the prediction residual. The information source decoding unit 109 decodes a voice signal from these codes and outputs it.

At this time, the error rate monitoring circuit 110 determines the error rate state of the wireless transmission path (communication path) 105 from the information obtained by the error correction decoding circuit 108. The result is transmitted to the control circuit 111, and the error correction coding circuit performs a correction process on the information transmitted from the B side to the A side according to the error rate. The information of the control circuit 111 is also included in the information. The error rate information transmitted from the B side is transmitted to the A side control circuit 211, and the A side error correction coding circuit 102 is transmitted.
Used to control the.

In the system of FIG. 4, the transmission information, that is, the encoding process is adapted according to the error rate state of the transmission line (communication line).

[0016]

In the first prior art example, the information protected by error correction can be completely corrected if the error is below the correction capability, but no processing is performed on the unprotected information. There is a problem that sound quality deterioration occurs when the unprotected and unprotected code is incorrect.

In the second prior art example, the error state of the communication path is monitored and the coding process of the transmission information is adapted. However, the voice coding process after detecting the error is adapted. Therefore, there is a time delay in dealing with an error. Also,
Since adaptive processing is incorporated in the encoding itself, there is a problem that the system becomes complicated.

The present invention was devised in view of such circumstances, and an object thereof is to suppress sound quality deterioration due to an error of an unprotected code while simplifying the configuration.

[0019]

A speech decoding apparatus according to the present invention is an error correction decoder in a speech encoding / decoding system comprising speech encoding / decoding means and error correcting means for speech code. Error rate calculation means for estimating the error rate related to the unprotected information based on the error bit number for the protected information obtained in step 1, and the correction necessary for the speech decoding process according to the estimated error rate. It is characterized in that it is provided with an error correcting means for performing.

[0020]

When the error correction decoding process is performed on the received code, if the error rate is equal to or lower than the correction capability, it is possible to decode an error-free code for the information protected by the error correction and to protect the code. Get the number of error bits in the information. The error rate calculation means estimates the error rate of unprotected information on the basis of the number of error bits, transmits the estimated error rate to the error correction means, and the error correction means decodes the speech based on the estimated error rate. Since the necessary correction processing is instructed to the means, the sound quality deterioration due to the code error of the information not protected by the error correction is suppressed.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a speech decoding apparatus according to the present invention will be described in detail below with reference to the drawings. For convenience of explanation, the application in the voice communication field will be described, but the same can be applied to the storage of a voice signal in a semiconductor memory or the like.

FIG. 1 is a block diagram showing the configuration of a voice encoding / decoding system in which the voice decoding apparatus according to the embodiment is used.

A left side is a transmitter and a right side is a receiver with the communication path 304 as a boundary. The configuration on the transmitter side is basically the same as that described in the first prior art example. That is, the transmitter is composed of the following elements. A speech coder 301 that compresses a speech signal and outputs the speech code by classifying the speech code according to importance, and an error correction encoder 30 that protects a speech code of high importance by error detection / correction.
2 and a multiplexer 303 that multiplexes the error-correction code output from the error-correction encoder 302 and the error-correction code and the unprotected code directly output from the speech encoder 301, and transmits the multiplexed signal to the communication path 304. It consists of and.

The receiver is composed of the following elements:
A demultiplexer 305 that separates the multiplexed code received from the communication path 304, an error correction decoder 306 that decodes the code protected by error correction, and a received code in the received code in the error correction decoding process. Error rate calculator 307 (error rate calculating means) for calculating an estimated error rate based on the number of error bits
And an error protection processor 308 (error correction means) for instructing a voice decoding process to perform a correction process in accordance with the estimated error rate, and a code output from the error correction decoder 306 and subjected to the error correction decoding process and multiplexed. The audio decoder 309 is configured to decode the audio signal based on the code directly output from the separator 305 and the instruction from the error protection processor 308.

The operation on the transmitter side is the same as that of the conventional system. The operation on the receiver side will be described below. The received code is
The demultiplexer 305 separates the class 1 code protected by the error correction coding process and the unprotected class 2 code. The protected code is subjected to error correction decoding processing in the error correction decoding unit 306. At this time, if the error in the protected code is within the correction capability, complete error correction is performed and the number of bit errors in the protected code is known. Error correction decoder 306
Sends this information to the error rate calculator 307. The error rate calculator 307 estimates the error rate in the currently received unprotected code and sends it to the error protection processor 308. The error protection processor 308 corrects the speech decoding process for the speech decoder 309 based on the information of the estimated error rate of the currently received code, and when there is an error in the unprotected code, Suppress sound quality deterioration. The details will be described below.

A specific example of calculating the number of bit errors in the protected code in the error correction decoder 306 will be shown. For example, when BCH coding is used for error correction, the bit error position and the number thereof are obtained during the calculation of the error locator in the decoding process, so the number of bit errors is obtained in the normal decoding process. To be

As another example of calculating the number of bit errors, when the convolutional coding / Viterbi decoding process is used, if the basic path metric calculation means is used, the surviving path metric is used to detect the error. The number is obtained. An example of this operation is shown in FIG. 2 from "Code Theory" (Chapter 12, Imai, Corona). This is because the coding rate is 1/2,
FIG. 3 is a trellis diagram of a binary convolutional code in which a generator matrix is a mathematical expression, and G = [1 + D ² 1 + D + D ² ]. In the figure, the branch represented by the solid line corresponds to the information bit 0, and the broken branch corresponds to the information bit 1. Branches marked with "x" in the figure are truncated. The number attached to each state is the surviving path metric that reaches that state. Two tail bits are given to terminate the data.

In the case of this example, it can be seen that there is a 3-bit error in the received code (see contrast between code sequence and received sequence). The difference (that is, the number of errors) between the code sequence and the reception sequence can be known by the minimum path metric path of Viterbi decoding.

A specific example of calculating the estimated error rate in the error rate calculator 307 will be shown. The error rate calculator 307 receives the number of bit errors from the error correction decoder 306 and estimates how many errors are in the unprotected class 2 code. The simplest example of a concrete estimation method is a method of directly using the number of bit errors at the current time calculated by the error correction decoder 306.

There is also a method of averaging the number of errors in a short time and calculating the short-time average error rate of the communication state. In the case of a communication path such as a car phone, errors are not only bit errors, but also burst errors due to the influence of the distance to buildings and base stations. In this case, it can be expected that bit errors at successive times have a short-term correlation. By using this, a stable error rate averaged over time can be obtained by using, for example, a moving average of bit error rates at several times in the past, rather than using information at only one time.

The estimated error rate is obtained by adding the number of errors at each time and dividing by the number of samples.

Alternatively, when the error pattern is not bit error-like, for example, when it is a fading or burst-like error pattern, it is possible to predict the temporal error pattern. Although the method described above is merely the estimation based on the average of past information, here, the estimation method is performed using a predicted value. For example, if the time series of the error rate has a first-order autocorrelation value of 0.8, a is an appropriate constant less than 1, and the number of estimated errors (t) = 0.8 × the number of estimated errors (t−1).
+ A x current error count is calculated. In this way, the estimation is performed from the error time series of the received code using the linear prediction method.

A specific example of the correction process performed by the error protection processor 308 on the speech decoder 309 will be described. The error protection processing unit 308, based on the estimated error rate received from the error rate calculation unit 307, performs error correction processing for suppressing sound quality deterioration when there is an error in a code that is not protected by the speech decoding unit 309. Instruct. The correction processing is roughly classified into the following two types. The first method is to correct the received code, and the second method is to correct the speech decoding process itself.

The first method is to add a correction to the voice code so as to reduce the influence of an error. For example, in the case of a code expressing the voice power, there is a method of making the code so that the output level becomes smaller as the error rate becomes higher, so that the noise becomes less noticeable.

The second technique is to correct the speech decoding process itself without modifying the received code, and this technique can perform various correction processes. As an example of the correction, as described above, there is a method of directly correcting the voice output level by adapting to the error rate, and based on the corrected voice power = original voice power × (1-estimated error rate) Will be corrected. Here, the estimated error rate is a value in the range of 0 to 1 obtained by the error rate calculator 307.

As another example, there is a method of weakening the resonance characteristic of the filter when a linear prediction filter is provided as in CELP speech coding. In addition, CELP
For speech coding, see "Code-Excited Linear Predi
ction (CELP): High-Quality Speech at Very Low Bit Ra
tes ", MRSchoeder and BSAtal, Proc.IEEE Int.Conf.
on Acoustics, Speech and Signal Processing, pp.937-9
40,1985 ".

As a method of weakening the resonance characteristic of the linear prediction filter, there is a method of performing bandwidth expansion correction. For this, see, for example, “A Class of Analysis-by-Synthesis Pre
dictive Coders for High Quality Speech Coding at R
ates Between 4.8 and 16 kbits ", P.Kroon and EDF.Dep
rettere, IEEE Journal on Selected Areas in Commun.V
ol.6 No.2, pp.353-363, 1988 ". The following can be mentioned as an example of processing. Corrected A [i] = A [i] × (1-estimated error rate) ⁱ , where the linear prediction coefficient of the i-th order is A [i] ⁱ Note that errors exceeding the correction capability of the error correction decoder 306 However, as described in the first related art, error protection processing such as squelch is performed. In this case, the error correction decoder 306 outputs a large number of error bits to the error rate calculator 307.
To transmit.

[0038]

As described above, according to the present invention, when an error occurs in the code of information that is not protected by error correction,
Deterioration of the voice quality can be suppressed by applying a necessary correction to the voice decoding process based on the estimated error rate obtained from the number of error bits in the protected information. In terms of configuration, it is a treatment for the error correction decoder and the speech decoder and does not affect the encoder side, so a relatively simple configuration improvement is required.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a voice encoding / decoding system in which a voice decoding device according to an embodiment of the present invention is used.

FIG. 2 is a diagram for explaining an example of an operation of obtaining the number of error bits in the speech decoding device according to the embodiment.

FIG. 3 is a block diagram showing a configuration of an error correction coding unit according to a first conventional technique.

FIG. 4 is a block diagram showing a configuration of a speech coding method according to a second conventional technique.

[Explanation of symbols]

 301 ... Voice encoder 302 ... Error correction encoder 303 ... Multiplexer 304 ... Communication channel 305 ... Demultiplexer 306 ... Error correction decoder 307 ... Error rate calculator (error Rate calculation means) 308 ... Error protection processor (error correction means) 309 ... Speech decoder

Claims (1)

[Claims] 1. In a voice encoding / decoding system comprising voice encoding / decoding means and error correcting means for a voice code, the number of error bits for protected information obtained by an error correction decoder is set. A voice characterized by comprising an error rate calculation means for estimating an error rate related to unprotected information based on the error rate, and an error correction means for performing a necessary correction for a voice decoding process according to the estimated error rate. Decoding device.