JPS62278600A - Synthesization of variable length frame type code encoder - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention [Field of Industrial Application] The present invention relates to a method of synchronizing a variable length frame type code decoding device &<
In particular, the present invention relates to a method for synchronizing a variable-length frame type code/decoding device that significantly improves call delay in synchronization processing.

[Conventional technology]

Coding/decoding that analyzes input speech at the basic analysis frame period, extracts spectral envelope parameters and sound source information as speech parameters, encodes this, and sends it to the decoding side via a transmission path to synthesize the original input speech. oxidation equipment is well known.

In this case, the spectral envelope parameter indicating the macroscopic characteristics of the input speech spectrum is L P C (LinearPr
diction coding, linear predictive analysis)
Coefficients or their derived forms are used, including α parameters, J, K parameters, or LSP (
Various coefficients such as Line Spectrum Pa1rs and Line Spectrum Pairs are used. In addition, as sound source information that indicates the microscopic characteristics of the input speech spectrum, the pulse frequency corresponding to the pitch frequency and white noise are necessary for encoding and decoding by modeling them corresponding to voiced/unvoiced information. Pitch frequency, voiced/unvoiced, sound source amplitude, etc. are used,
Other types of code that utilize the waveform information of the sound source include multipulse and residual waveform.

It may also be decrypted.

A variable-length frame type code/decoder is a variable-length frame type code/decoder that performs multiple analyzes instead of faithfully transmitting spectral envelope parameters in analysis frame units among audio parameters to be sent from the coding side to the decoding side in the code/decoding device described above. By selecting a representative analysis frame representing the frames, and transmitting the spectral envelope parameter of this representative analysis frame and the number of analysis frames designated as having the same spectral envelope parameter as this representative analysis frame, so-called repeat bits. This realizes a significant reduction in transmission bit rate.

Although there are various methods for constructing a variable length frame in such a variable length frame code decoding device, the following are basically used.

That is, an optimum approximation relationship is set for each section consisting of a preset number M of analysis frames to optimally approximate the section spectrum envelope. Examples of functions used in this piecewise optimal approximation include rectangle 5&lt1, etc., and the settings are such that the approximate rectangular envelope is divided such that the sum of the distances between spatial vectors is the minimum according to the spectral envelope parameter for each analysis frame. It is done in the form of a ``head''.

A plurality of i analysis frames constituting the approximate rectangle thus obtained are assumed to be seven representative frames representing the division. The above-mentioned i pieces are arbitrarily set corresponding to the respective spectral envelopes for each section, and such optimal approximation itself is based on the total number of interspace vectors based on the spectral envelope parameters for each analysis frame, including the analysis order. The sum of the hit vector distances is used as an evaluation measure, and the combination of analysis frames that minimizes this is called Dr (Dynamic Pr
ograming.

It is also well known that it can be easily determined using dynamic programming (dynamic programming).

Now, the spectral envelope parameters obtained in the variable length frame format are encoded together with the sound source information in a predetermined format and transmitted to the decoding side, but in this case,
Seven-rem synchronization is required between the encoding side and the decoding side in data transmission and reception.

As this synchronization format, either independent synchronization or forced synchronization is often used, but in either case, one segment that makes up the variable length frame is treated as the basic frame in the synchronization process. .

In other words, in independent synchronization, in order for the decoding side to perform synchronization control to eliminate the slip caused by the clock frequency difference with the code side, it is necessary to accumulate data in units of one section, and in case of synchronization deviation, this one section Synchronization is recovered through frame updating due to unit data deletion, and clock generation is provided from the encoding side to the decoding side even in forced synchronization.
distribution.

At present, data is accumulated in units of one section in order to eliminate the effects of disturbances due to transmission, etc., and this is also done on the premise of frame synchronization by one section shift.

[Problem that the invention seeks to solve]

The synchronization method of the conventional variable-length frame type code decoding device described above is to accumulate data in units of one section, regardless of independent synchronization or forced synchronization, and achieve frame synchronization using one section as a Z'lA frame unit. If this method is used, there is a problem in that a significant call delay is unavoidable compared to a normal code/decoder that performs frame synchronization on an analysis frame basis.

This call delay is due to the fact that one segment of the variable length frame is several hundred meters long.
This is obvious considering that it is not uncommon for the amount to be around 5EC.

SUMMARY OF THE INVENTION An object of the present invention is to eliminate the above-mentioned drawbacks, and to provide a synchronization method for a variable-length frame type code decoding device, which is sufficiently shorter than one section, for example, a basic analysis frame or a time slot corresponding thereto. It is an object of the present invention to provide a method for synchronizing a variable-length frame type code decoding device, which significantly improves call delays caused by synchronization processing by providing means for achieving frame synchronization only by accumulating data.

[Means for solving problems]

The method of the present invention is a synchronization method for a variable-length frame code decoding device that uses independent synchronization or forced synchronization for synchronization with the encoding side. A voiced section or a silent section for each variable-length frame segment, or a time length of the voiced section and silent section is adaptively controlled in accordance with the degree of synchronization difference between the coding side and the decoding side, and the encoding side and the decoding side are The device is equipped with synchronization maintaining means for maintaining synchronization with the other side.

〔Example〕

Next, the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention, which consists of an encoding side 1 and a decoding side 2.

The encoding side 1 further includes an A/D converter 11° window processor 12. LSP analyzer13. Sound source information analyzer 14. Variable length 7 frame selector 15. It is configured to include a clock generator 16, a multiplexer 17, and the like.

The decoding side 2 also includes a demultiplexer 21° frame synchronizer 22 . Fixed length framer 23°Mesosu 24. Adaptive controller 25. Modified message generator 26. It is configured to include a clock generation circuit 27, a speech synthesizer 28, and the like.

The embodiment shown in FIG. 1 uses independent synchronization as the synchronization method and uses LSP coefficients as the spectrum envelope parameter, but this is a forced synchronization method and uses other oLPC coefficients. There is no problem with this, and basically it can be implemented in almost the same way.

Now, the input audio is supplied to the A/D converter 11, and the high cutoff frequency is 3.4 KHz OLPF (Low P
After removing unnecessary high frequencies through a 5s1;'1lter), the signal is sampled at a sampling frequency of 8KH2 and digitized with a 12-bit quantization step.

When the window processor 12 receives the quantized audio signal from the A/D frequency converter 11, the window processor 12 temporarily stores one preset block worth, for example, 30m5Ec (240 samples) each in a memo IJIc, and stores the preset settings therein. The window processing to cut out by multiplying the Hamming function of the window function is 10m5E.
This is carried out every analysis frame period of C. Therefore, the basic analysis frame length is 10m5EC.

The output of the window processor 12 is supplied to an LSP analyzer 13 and a sound source information analyzer 14.

The LSP analyzer 13 performs LPC analysis on data for each input analysis frame using a known method to obtain a predetermined 8th-order α
The parameters are extracted and then used to derive LSF coefficients as S-dimensional space vectors. In this embodiment, an even number is selected as S. The C8P coefficient is the following (1)
It is represented by P in the formula.

? (Pl + "2 +...Ps)...
・・・・・・・・・・・・・・・・・・(1) P1 to Ps shown in equation (1) are expressed in the form of a set of resonance frequencies obtained based on the open/closed state of the vocal tract. It is determined by frequency domain parameters.

LAP for each analysis frame obtained by L8F analyzer 13
The coefficients are supplied to a variable length frame selector 15.

The variable length frame selector 15 performs variable length frame selection based on piecewise optimal function approximation in the following manner for a series of spectral envelope parameters input in units of analysis frames.

That is, in this embodiment, one section is an analysis frame section with inheritance<20, that is, 200 m5 EC, and piecewise optimal function approximation is performed on this section by rectangular approximation. In this case, it is of course possible to use an approximation function such as a polygonal line approximation function instead of the rectangular function as the approximation function.

Now, it is assumed that i pieces out of 20 analysis frames included in each category are selected as representative analysis frames. Here, i is a number arbitrarily selected from 1 to 20. The LAP coefficient vectors belonging to this representative analysis frame are used to approximately represent the LSF coefficient vectors belonging to other analysis frames in this division. i is selected so that . Distortion in this case is basically obtained using the vector distance of the coefficients as described above as a measure, and the DP method is used based on the viewpoint that the minimum distortion is the one with the minimum sum of vector distances between coefficients for each analysis frame. can be easily determined.

In this way, representative analysis frames of the next i i HI3 are selected for each division, and the analysis frames to be represented by these i representative analysis frames are also determined, and are supplied to the multiplexer 17 as LAP coefficient data having a variable length frame structure. be done.

The sound source information analyzer 14 uses the quantized audio signal for each analysis frame received from the window processor 12, obtains the pitch period, voiced/unvoiced status, and sound source amplitude using a well-known analysis method, and analyzes these sound source information. Each frame is supplied to the multiplexer 17.

The clock generator 16 is necessary for digital processing from the A/D converter 11 to the multiplexer 17 and provides a lock t1.

The multiplexer 17 sends the input LAP coefficient data and sound source information data to the decoding side 2 via the transmission path 100 as a multiplexed signal that is synthesized and encoded in a form suitable for transmission.
Therefore, information regarding the clock t1 is also included in this multiplexed signal and provided to the encoding side 2.

On the decoding side 2, the multiplexed signal provided from the multiplexer 17 on the encoding side 1 is received by the demultiplexer 21,
Based on the spectrum of this time series data, clock t1
is extracted, the multiplexed signal is separated using this clock t1, and then decoded and restored to the input state of the multiplexer 17 on the encoding side 1.
The P frequency data and the sound source information data provided in units of analysis frames are supplied to the frame synchronizer 22. Further, the clock 11 extracted by the demultiplexer 21 is transmitted to the frame synchronizer 22. Fixed length framer 23. Memory 24. The corrected message generators 26 and the like are each supplied with the corrected message generators 26 and the like.

The frame synchronizer 22 receives the clock t thus provided.
Using the information regarding LAP coefficient data, frame synchronization is performed between the divisional frame to which the LAP coefficient data was sent and the analysis frame to which the sound source data was sent, and these data are received and supplied to the fixed-length frame generator 23.

The frame synchronizer 23 also performs frame synchronization while extracting repeat bits from the received LSP coefficient data and data related to electric kabino from the sound source data, and supplies these to the adaptive controller 25. The above-mentioned repeat bit is data regarding the number of analysis frames represented by each of the i rows of representative analysis frames selected for each section and the order within the section.

Fixed length framer 23 receives clocks 1 . Under the drive of , representative analysis frames and repeat bits are used to perform interpolation between representative analysis frames to restore the specified state of LSF coefficients for each fixed frame. That is, the output state of the LSP analyzer 13 is restored.

In this way, the LSP coefficients converted into fixed-length frames are separately stored in the memory 24 as audio parameters in analysis frame units together with the sound source information provided from the frame synchronizer 22 in analysis frame units.

The write clocks for storing data in the memory 24 are supplied from the demultiplexer 21 and are clocks 1. The clock t2 shown by the dotted line supplied from the clock generator 27 on the decoding side 2 is used as the clock for reading this and supplying it to the speech synthesizer 28. These two clocks are used to maintain synchronization between the encoding side and the decoding side in a mutually independent synchronization format. Although this type of independent synchronization is superior in terms of ensuring clock stability on the encoding side, flexible system configuration, and efficient failure handling, it also suffers from call delays caused by the time difference between the two clocks. Inevitable. This will be explained in detail as follows.

Now, if the clock t2 on the decoding side 2 is shifted in a faster direction than the clock t1 on the encoding side, the reading speed will be faster than writing to the memory 24, and eventually there will be no frame data to read. This will result in In addition, if the clock t2 deviates in a direction slower than the clock 11, the VC
A situation occurs in which a write command for the next 7 frames of data is issued. In such a case, while monitoring the state of the deviation, if it exceeds a certain tolerance value, the writing and reading of the next analysis 7 frames will be interlaced and shifted, and the code will be changed.
Frame synchronization between decoding sides is ensured.

However, in a variable length frame type code/decoder,
Frame synchronization adjusted in this manner processes one segment as a basic frame unit, and the call delay due to this is significantly increased compared to a normal code/decoder.

In this embodiment, the above-described frame synchronization correction is performed in analysis frame units that are sufficiently shorter than one segment, thereby achieving a significant improvement in call delay.

Each time the fixed length framer 23 receives data regarding an adjustment frame from the adaptive controller 25, it deletes the relevant analysis frame from the analysis frames to be output to the memory 24. This adjustment frame is used for stationary voiced sounds whose time length is essentially indeterminate and therefore has a sufficiently small influence on the reproduced sound quality, excluding consonants or consonant-vowel coexistence parts that are in the segment and have a strong influence on the reproduced sound quality. and a silent part, or a pause part in the voice that is effectively equivalent to the silent part, and the timing is such that it is performed every time a modified message is received from the modified message generator 26.

By the way, in the spectral envelope parameter converted into a variable length frame, the time length of the analysis frame group specified by the representative analysis frame is long, so the influence of one analysis frame deleted to maintain frame synchronization is relatively One of the sections that requires less is the voiced stationary section υ, and the other is the silent section that includes pauses. When these voiced sound stationary parts and silent parts are evaluated from the viewpoint of voice power, the former clearly has high power and the latter has low power. Which analysis frame is to be deleted is selected from the seven analysis frames specified by the repeat bit using the power bit provided by the frame synchronizer 22 as follows.

The modified message generator 26 constantly receives the clock 11 and the clock t2 and monitors the time difference ±Δt between the clocks. When this time deviation ±Δt exceeds a certain tolerance value, the modified message generator 26 outputs a modified message that causes the adaptive controller 25 to output adjustment frame data specifying the analysis frame to be deleted from the classification.

When the adaptive controller 25 receives the correction message, it determines the temporally nearest analysis frame to be deleted based on the electric pattern and the repeat bit, outputs this information as adjustment frame data, and maintains synchronization in units of analysis frames. Let's do it.

The speech synthesizer 28 receives the clock t2 and the memory 23
The spectral envelope parameters and sound source data are supplied in units of analysis frames.

The speech synthesizer 28 uses LSP coefficients as filter coefficients of the LAP synthesis filter, drives this filter with a sound source signal reproduced from sound source data to generate a digital sound signal, and converts this into a D/A converter to generate analog sound. After concatenating the signal, unnecessary high-frequency components of 3.4 KHz or higher are removed using an LPF to obtain output audio.

Thus, a variable length frame code with segmentation units.

While performing decoding, frame synchronization maintenance can be processed on an analysis frame basis.

Note that in this embodiment, the time length of the section specified by the representative analysis frame and the time length of the silent section among the spectral envelope parameters are jointly used for adaptive control, but each of these is adaptively controlled independently. However, there is no problem at all.

In addition, in this implementation, the analysis frame is used as a basic frame for synchronization, which is a sufficiently shorter period than one division, but it is clear that this can be easily implemented even when the number of samples of approximately the same time length is targeted. be.

〔Effect of the invention〕

As explained above, according to the present invention, in the synchronization method of a variable length frame type encoding device that uses independent synchronization or forced synchronization for synchronization between the encoding side and the decoding side, In particular, by providing a flat part that adaptively controls the interval length or these two time lengths in accordance with the degree of synchronization between the encoding side and the decoding side, it is possible to significantly reduce call delays caused by correction of synchronization deviation. There is an effect that it is possible to realize a synchronization method for a variable-length frame code decoding device that reduces synchronization.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention. 1...Encoding side, 2...Decoding side, 1
1... A/D converter, 12... Window processor, 13... LAP analyzer, 14...
...Sound source information analyzer, 15...Variable length frame selector, 16...-Clock generator, 17...
... multiplexer, 21 ... demultiplexer, 22 ... frame synchronizer, 23 ...
...Fixed length framer, 24...Memory, 2
5...Adaptive controller, 26...Modified message generator, 26°27...Clock generator, 28...Speech synthesizer, 100... ...Transmission line. Agent Patent attorney Susumu Uchihara Procedural amendment (voluntary) Relationship to the Showa 1962-10 Incident Applicant 5-33-1 Shiba, Minato-ku, Tokyo (423) NEC Corporation Representative Tadahiro Sekimoto 4 , Agent & Column 6 of “Detailed Description of the Invention” of the specification subject to amendment, Contents of amendment

Claims (1)

[Claims] In a synchronization method for a variable-length frame type code/decoder that uses independent synchronization or forced synchronization to synchronize an encoder and a decoder, The coding side and the decoding process are performed while adaptively controlling the voiced period or the silent period for each variable-tone frame segment, or the time length of the voiced period and the silent period according to the degree of synchronization difference between the coding side and the decoding side. 1. A method for synchronizing a variable-length frame type code decoding device, comprising: synchronization maintaining means for maintaining synchronization with a coding side.