JPH0351900A - Error processing system - Google Patents

Abstract

PURPOSE:To reduce the unnaturalness of a reproduced frame and to improve the sense of hearing by generating a code word group as a candidate for an error frame, and selecting the reproduced waveform having the shortest distance to a reference waveform as a reproduced waveform which is one or several pitches precedent as the reproduced waveform of the error frame in a reproduced waveform group composed from the code word group. CONSTITUTION:An error detection part 1 detects the bit error of a received code word and a candidate code word generation part 2 generates the code word group as a candidate for a correct code word as to the received code word within a range corresponding to the number of error bits detected by the error detection part 1. A waveform generation part 3 generates the candidate reproduced waveform group corresponding to the candidate code word group generated by a candidate code word generation part 2 by using the pitch information of the received data 3. A storage part 4 is stored with the reproduced waveform which is one or several pitches precedent as the reference waveform and an evaluation part 5 selects the waveform closest to the reference waveform read out of the storage part 4 in the candidate reproduced waveform group generated by the waveform generating part 3. Therefore, there is not deterioration in sound quality and information on variation from the last frame is utilized. Consequently, the unnaturalness of the reproduced frame is reduced and the sense of hearing is improved.

Description

[Detailed Description of the Invention] [Summary] An object of the present invention is to provide an error processing method in a decoder of a high-efficiency audio encoding method, which can reduce unnaturalness in reproduced frames and improve audibility. In a decoder using a high-efficiency audio coding method that collects reproduced waveforms from received codewords for each frame, an error detection unit that detects bit errors in received codewords and a corresponding number of detected error bits are provided. a candidate codeword generation unit that generates a codeword group that is a correct codeword candidate for the received codeword within a range that corresponds to the received codeword; and a candidate reproduced waveform that corresponds to the generated candidate codeword group using pitch information of the received data. comprising a waveform generation unit that generates a group of waveforms, a storage unit that stores a reproduced waveform from one or several pitches earlier as a reference waveform, and an evaluation unit that selects a waveform closest to the reference waveform from the candidate reproduced waveform group; The selected waveform is output as a reproduced waveform of the error frame.

[Industrial Application Field] The present invention relates to an error processing method in a decoder of a high-efficiency speech encoding method, and for a frame in which an error has been detected, reproduces a waveform that is closest to a reference waveform among a group of candidate waveforms. The present invention relates to an error handling method used as an error handling method.

In high-efficiency audio encoding, especially in the case of audio decoders that are applied to transmission paths with low reliability, errors are detected for each frame of the received codeword, and the frame in which an error is detected is detected. Some kind of error processing is performed for these errors to reduce the effect on the reproduced audio.

It is desired that such error processing should be able to reduce the unnaturalness of the reproduced frames and improve the audibility.

[Conventional technology]

In high-efficiency audio encoding, several methods as shown in FIG. 7 have conventionally been used to process frames in which errors have been detected.

In FIG. 7, (a) shows a processing frame.

The reproduced sound for this has a waveform as shown in (b) when there is no error in the received code word. On the other hand, if there is an error in the received code word and error processing is not performed, the reproduced sound corresponding to the error frame will have a degraded waveform as shown in (C).

(d) shows an example of a conventional reproduction sound error processing method,
The error frame is replaced with the waveform of the preprocessed frame. The replacement in this case is Linear Prediction (LPG)
This can also be done by replacing the combination coefficients in the combination method.

(e) shows another example of the conventional reproduction sound error processing method, in which an error frame is silenced.

It is known that by performing these error processes, a somewhat more natural hearing sensation can be obtained compared to the unprocessed reproduced sound shown in (C).

[Problem B to be Solved by the Invention] However, the conventional error processing method has the following problems.

The method of silencing shown in FIG. 7(e) has a problem in that especially in a transmission path where the error rate is high, the sound becomes intermittent, resulting in deterioration of sound quality, which is not pleasing to the listener.

Furthermore, in the method of replacement using the previous frame waveform shown in FIG. 7(d), no information on changes from the previous frame to the current frame is contributed. In other words, as shown in FIG. 7 (b), the audio signal often changes gradually while repeating at a pinch cycle, but if it is replaced with the previous frame waveform, the fluctuations between frames are ignored.
After all, it is inevitable that an unnatural hearing sensation will occur.

The present invention is an attempt to solve the problems of the prior art, and is to perform error processing on frames in which an error is detected in the received code word in a decoder using a high-efficiency speech encoding method. It is an object of the present invention to provide an error processing method that reduces unnaturalness in reproduced frames and improves hearing sensation.

[Means to solve the problem]

As shown in FIG. 1, the present invention uses an error detection unit 1, a candidate codeword It includes a generation section 2, a waveform generation section 3, a storage section 4, and an evaluation section 5, and outputs the waveform selected by the evaluation section 5 as a reproduced waveform of an error frame.

The error detection unit 1 detects bit errors in received code words.

The candidate codeword generation section 2 generates a group of codewords that are candidates for correct codewords for the received codeword within a range corresponding to the number of error bits detected by the error detection section 1.

The waveform generator 3 uses the pitch information of the received data to generate a group of candidate reproduced waveforms corresponding to the group of candidate codewords generated by the candidate codeword generator 2.

The storage unit 4 stores a reproduced waveform from ■ to several pitches earlier as a reference waveform.

The evaluation unit 5 evaluates the reference wave read out from the storage unit 4 from the candidate reproduction waveform group generated by the waveform generation unit 3 . The waveform closest to is selected.

[Effect]

FIG. 2 conceptually shows error processing in the present invention, where A indicates the code word area and B indicates the reproduced waveform area. In FIG. 2, it is assumed that the decoder to which the method of the present invention is applied is capable of detecting m-bit errors.

The received codeword of the previous frame is now C, and the received codeword of the current frame containing an error is C. Then, when m bit errors occur, the code word C 1, C 2, -”+ C mrn, which is a candidate for the received code word, has one bit around the code word C when there is no error. - They can exist as shown within a distance corresponding to several meters.

code word c. , c as the combined waveform from s. , s is obtained. In addition, candidate reproduced waveforms s I, s . ,・1,s■ is obtained. Using the reproduced waveform for one pitch period just before (from E to several pitches ago) as the reference waveform srar, these candidate reproduced waveforms are compared with the reference waveform Sref, and the most A waveform that is close in distance is selected as a reproduced waveform.

FIG. 3 explains the selection of reproduced waveforms.

In Fig. β, (a) shows a processed frame, and it is assumed that the i frame is an error frame. The reproduced sound corresponding to this becomes abnormal after the error frame, as shown in (b).

(C) shows the pitch period in the reproduced sound, where p is the final pitch position at the pitch position of the normal frame (indicated by the numbers), and the pitch position of the error frame estimated from the pitch position of the normal frame. (Indicated by ↑)
Let P′ be the pitch period in this part, and let T be the pitch period in this part.

Reference waveform Sr* which is the reproduced waveform for one pitch period immediately before
For the pitch position p of f, the aforementioned candidate reproduced waveform "I+" in the error frame 2. ...+SIIf
The pitch positions W p ' of fi are made to match. At this time,
Let the length of each waveform be the pitch circumference T. Therefore, the time from the beginning of the waveform to the pitch position is all a constant time L.

Among these, the reference waveforms s, . ．． The waveform closest to f (S4
) is extracted as a candidate waveform (=s., t) of the error frame.

In the present invention, this candidate waveform S. pt is used as the reproduced waveform of the error frame. In this way, the closest reproduced waveform to the previous one fundamental period (pinch period) with a strong correlation is selected as the reproduced waveform, so there is less deterioration in sound quality compared to conventional methods, and there is less variation from the previous frame. The information will be contributed.

In FIG. 7, (f) shows the reproduced waveform of the method of the present invention in comparison with the reproduced waveform of the conventional method, and compared to the conventional method, the waveform from a normal frame has the least deterioration. It has been shown that

(Example) FIG. 4 shows an example of the present invention.
12 is an error correction/detection section, 12 is a candidate coefficient generation section, 13 is a combination section, 14 is a buffer, and 15 is an evaluation section.

FIG. 4 shows a case where pitch information is transmitted from the encoder side (transmission side) together with waveform combining coefficients.

? The error correction/detection unit 11 performs error correction and error detection in the code word input using a normal error correction/detection method.

Suppose now that an m-bit error is detected in the code word of an i-frame. Candidate coefficient generation unit l2 generates candidate codeword groups C + . C 2+'
−′+ C me (mn+=2″), L
Generate PC synthesis coefficients. The synthesis unit l3 uses the coefficients of the candidate code words to generate reproduction waveform candidates sl, s,...
, S■ are synthesized. On the other hand, the reproduced waveform of the previous frame is stored in the buffer 14 as a reference waveform. A reference waveform is read from the buffer 14 according to the pitch position information transmitted from the transmitting side.

By matching the pitch position of this reference waveform with the pitch position of the candidate waveform, waveform matching for l pitch periods is performed to find a difference. The evaluation unit 15 selects the candidate waveform closest to the reference waveform as the waveform candidate for the error frame. The selected waveform is output as playback sound.

FIG. 5 shows another embodiment of the present invention, in which the same parts as in FIG. 4 are designated by the same numbers, and 16 is a pitch extraction section.

In the embodiment shown in FIG. 5, an error correction/detection section 11, a candidate coefficient generation section 12. The operation of the gathering section 13 is the same as that of the embodiment shown in FIG.

In the embodiment of FIG. 5, pitch information is not transmitted from the transmitter. The pitch extraction unit l6 extracts the pitch position from the waveform of the previous frame stored in the buffer 14. A reference waveform is read from the buffer l4 using this pitch position as a reference.

The pinch position of this reference waveform is matched with the pitch position of the candidate waveform, and waveform matching for one pitch cycle is performed to find a difference.

The evaluation unit 15 selects the candidate waveform closest to the reference waveform as the waveform candidate for the error frame, and the selected waveform is output as reproduced sound.

FIG. 6 shows still another embodiment of the present invention, in which CEL
P (Code-Excted Linear Pr
The case is shown in which the method is applied to a decoder using the ``edic'' method.

In FIG. 6, the same parts as in FIG. 4 are indicated by the same numbers, and 2L22 is a candidate generation section, 23 is a codebook, 24 is a pitch reproduction section, and 25 is an LPG combination section.

The error correction/detection unit l1 performs error correction and error detection in the codeword input, and when an m-bit error is detected in the codeword of an i frame, the candidate generation units 21 and 22 generate a codeword without an error. , the candidate codeword group CI within the distance determined by the number m of bits including the received codeword
+C2+'-'+C1 ("=2'") A speech combination coefficient group is generated.

In this case, the candidate generating section 21 generates coefficients necessary for pitch reproduction, and the candidate generating section 22 generates coefficients necessary for LPC combining.

On the other hand, the codebook 23 stores a plurality of types of noise patterns that serve as sound sources in LPG combination, and are read out according to specifications from the encoder side (transmission side). The pitch reproduction unit 24 reproduces the pitch of the noise read from the codebook 23 using the coefficients from the candidate generation unit 2l.

The LPG synthesis section 25 performs LPC synthesis on the pitch-reproduced noise using the coefficients from the candidate generation section 22, and generates a reproduction waveform candidate S. SZ+'-'+''1 is read as a signal. On the other hand, the reproduced waveform of the previous frame is stored in the C/F14 as a reference waveform.

A reference waveform is read out from the buffer 14 according to the pitch position information reproduced by the pitch reproduction section 24, and the pinch position of this reference waveform is matched with the pitch position of the candidate waveform, thereby reproducing the waveform for one pitch period. Perform matching and find the difference. The evaluation unit 15 selects the candidate waveform closest to the reference waveform as the waveform candidate for the error frame.

The selected waveform is output as playback sound.

〔Effect of the invention〕

As explained above, according to the present invention, a candidate code word group is generated for a frame in which an error has been detected (an error frame), and one or several pitches of the reproduced waveform group synthesized from the code word group are generated. Since the one closest to the reference waveform, which is the previous reproduced waveform, is selected as the reproduced waveform of the error frame, the one close to the previous reproduced waveform with a strong correlation is output as the reproduced waveform. Therefore, according to the present invention, it is possible to reduce the deterioration in sound quality in reproduced audio compared to the conventional method, and also to make the fluctuation information from the previous frame contribute to the reproduced audio, so that the reproduced frame has less unnaturalness. , can improve hearing.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the principle structure of the present invention, Fig. 2 is a diagram conceptually showing error processing in the present invention, Fig. 3 (a),
(b), (C) are diagrams explaining the selection of reproduced waveforms, Figures 4 to 6 are diagrams each showing an embodiment of the present invention, and Figures 7 (a), (b), (c). ．． (cl),
(e) and (f) are diagrams showing conventional error processing methods. 1 is an error detection section, 2 is a candidate code word generation section, 3 is a waveform generation section, 4 is a storage section, and 5 is an evaluation section.

Claims (1)

[Claims] In a decoder using a high-efficiency audio coding method that synthesizes a reproduced waveform for each frame from a received codeword, an error detection unit (1) detects bit errors in a received codeword; a candidate codeword generation unit (2) that generates a codeword group that is a correct codeword candidate for the received codeword within a range corresponding to the number of error bits, and the generated candidate using pitch information of the received data; a waveform generation unit (3) that generates a group of candidate reproduced waveforms corresponding to the group of code words; a storage unit (4) that stores a reproduced waveform one or several pitches earlier as a reference waveform; An error processing method comprising: an evaluation unit (5) that selects a waveform closest to a reference waveform, and outputs the selected waveform as a reproduced waveform of an error frame.