JPH03186900A - Multipulse sound coding system - Google Patents

Abstract

PURPOSE:To reduce the influence of a synthetic signal generating an error by clearing a decoder synthetic filter internal data stored in a 2nd memory when clear information sent from a coder after multipulse decoding instructs to clear the decoder synthetic filter internal data. CONSTITUTION:When synthetic sound power is less than a prescribed threshold, the coder 10a clears the coder synthetic filter internal data stored in a 1st memory to prevent the influence of an error in a synthetic signal generated by the coder 10a due to a transmission bit error from being trailed up to the succeeding frame in the coder 10a. When the clear information instructs to clear the decoder synthetic filter internal data, the decoder 30a clears the decoder synthetic filter internal data stored in the 2nd memory. Thereby, the influence of an error in a synthetic signal generated by the decoder due to a transmission bit error can be prevented from being trailed to the succeeding frame.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a multi-pulse audio encoding system, and more particularly to a multi-pulse audio encoding system that prevents error propagation.

[Prior Art] FIGS. 5 and 6 are block diagrams showing an encoder and a decoder, respectively, of a conventional multi-pulse speech encoding system.

The audio signal processing operation of the multipulse audio encoding method is as follows.
This is done for each frame of a predetermined period of time, and in the conventional multi-pulse audio encoding system, the following method was used to smooth the connection between frames.

In the encoder, the voice synthesis data at the end of the current frame is stored in memory as encoder synthesis filter internal data, and in the next frame, an influence signal from the current frame is created based on this encoder synthesis filter internal data. However, the pulse search was performed after removing this influence signal from the original audio signal.

That is, in FIG. 5, the audio signal a input to the encoder 10 is inputted to the pulse search means 12 after an influence signal S, which will be described later, is subtracted by the adder 1. On the other hand, the synthesis filter coefficient calculated by the synthesis filter coefficient calculation means 1-5 based on the audio signal a is inputted to the pulse search means 12 via the quantization means 16 and the inverse quantization means 17. The pulse search means 12 performs a pulse search based on the manually inputted audio signal and the synthesis filter coefficients, and inputs the pulse data to the quantization means 13 . The quantization means 13 quantizes the pulse information and the synthesis filter coefficients, inputs the quantized pulse information C to the encoder 14,
The quantized synthesis filter coefficient d is input to the encoder 14. As a result, the encoder 14 outputs multi-pulse data f in which the pulse data C and the synthesis filter coefficient d are encoded.

Further, the pulse data dequantized by the dequantization means 18 is read out from the memory means (not shown) by the synthesis filter internal data loading means 19 and loaded into the encoder speech synthesis data, and the pulse data is sent to the speech synthesis means 20. It is synthesized into the audio signal e. The voice synthesis data at the end of the frame of the voice signal e is saved in a memory means (not shown) by the synthesis filter internal data save means 21. The influence signal creation means 23 combines the audio signal C input via the synthesis filter internal data saving means 21 and the input speech synthesis data read from a memory means (not shown) by the synthesis filter internal data load-1 stage 22. The influence signal created based on the above is output to the adder 11.

In addition, the decoder manually processes the multi-pulse data, synthesizes speech from the multi-pulse data, stores the speech synthesis data at the end of the current frame in memory as internal data of the decoder synthesis filter, and uses this decoding data in the next frame. Speech synthesis was performed based on the internal data of the synthesizer synthesis filter and pulse data.

That is, in FIG. 6, the encoder 1 shown in FIG.
Multi-pulse data f input from 0 to the decoder 30
is manually input to the decoder 31, separated into quantized pulse information C and quantized synthesis filter coefficient d, each dequantized by dequantization means 32 and 33, and then inputted from the synthesis filter internal data loading means 35. Together with the voice synthesis data, the synthesis filter 34 synthesizes the voice synthesis data a,) and outputs the synthesized voice data. Of the voice synthesis data aQ, the voice synthesis data at the end of the frame is stored in a memory means (not shown) by the synthesis filter internal data saving means 36.

[Problems to be solved by the invention] In the conventional multipulse speech synthesis method described above,
In both encoders and decoders, when errors occur in the speech synthesis data due to causes such as transmission bit errors, the effects of the erroneous synthesis signal may linger into subsequent frames.

[Means for Solving the Problems] According to the multi-pulse speech encoding method of the present invention, after multi-pulse encoding, if the power of synthesized speech is smaller than a predetermined threshold, the encoder clears the encoder synthesis filter internal data stored in the memory of the decoder, and sends clear information indicating whether to clear the decoder synthesis filter internal data and multipulse data to the decoder, After multi-pulse decoding, if the clear information sent from the encoder instructs to clear the decoder synthesis filter internal data, the decoder synthesis filter stored in the second memory I am trying to clear the filter internal data.

[Operation] When the power of the synthesized speech is smaller than a predetermined threshold, the encoder clears the encoder synthesis filter internal data stored in the first memory, thereby preventing the encoder from transmitting bit errors. This prevents the effects of errors in the composite signal from lingering in subsequent frames in the encoder.

If the clear information sent from the encoder instructs to clear the decoder synthesis filter internal data, the decoder clears the decoder synthesis filter internal data stored in the second memory. By clearing , the effects of errors in the decoder's composite signal due to transmission bit errors are prevented from lingering in subsequent frames in the decoder.

[Example] Next, the present invention will be explained with reference to the drawings.

FIGS. 1 and 3 are block diagrams of an encoder and a decoder in the multipulse speech encoding system of the present invention, and the same reference numerals as in FIGS. 5 and 6 indicate corresponding parts, respectively. , the explanation thereof will be omitted.

In FIG. 1 showing the encoder 1.0a, reference numeral 24 detects the power of the synthesized speech e input from the speech synthesis means 20 by comparing it with a threshold value, and detects the synthesis filter internal data clear information (hereinafter, clear), which will be described later. This is a synthesized speech power detection means that outputs g (referred to as information).

FIG. 2 shows an operation flowchart of the encoder shown in FIG. 1. In FIG. 2, the portion indicated by the dashed line is the operation of the portion corresponding to the multi-pulse audio encoding system of the present invention.

First, the synthesis filter coefficients are calculated by the synthesis filter coefficient calculation means 15 (step S], ), and the encoder synthesis filter internal data read from the memory means (not shown) is transferred to the influence signal from the synthesis filter internal data loading means 22. The audio signal is loaded into the creation means 23 (step S2), the influence signal is created by the influence signal creation means 23 (step S3), and the 'I3 signal is input to the adder 11, so that the audio signal manually input to the encoder 10a is The influencing signal is removed from a (original voice) (step S4). Pulse searching is performed by the pulse search means 12 based on the sound f2 from which the influence signal has been removed and the influence signal determined by the influence signal generating means 15, and pulse data is determined (step S5). Next, the speech synthesis means 2 manually inputs the pulse data obtained in step S5 and the encoder synthesis filter internal data loaded from the synthesis filter internal data loading means 19.
0, the synthesized speech signal e is output (step S6.
S7). The synthesized speech power detection means 24, which has received the synthesized speech signal e outputted in the previous step S7, detects the power of the synthesized speech signal e (step S8) and compares it with a threshold value (step S9). If the power of the signal e is smaller than the threshold value, clear the data in a memory means (not shown) that stores the internal data of the encoder synthesis filter.Step sio> Instructs to clear the internal data of the decoder synthesis filter. By inputting the clear information g, quantized pulse information C1, and quantized synthesis filter coefficient d output from the synthesized voice power detection means 24 to the encoder 14.a, the encoder 14a outputs multi-pulse data f1 ( Step S11>. Also,
In step S9, if the power of the synthesized speech signal e is equal to or greater than the threshold value, the voice synthesis data at the end of the frame of the synthesized speech signal e is stored in a memory means (not shown) (step 512), and the decoder synthesizes The clear information g output from the synthesized voice power detection means 24 which instructs not to clear the filter internal data, the quantized pulse information C1 and the quantized synthesized filter coefficient (1 is set to the encoder 1
4a and encode it to create multipulse data f1
(Step S]-1).

In FIG. 3 showing the decoder, 36a is a decoder synthesis filter stored in a memory means (not shown) according to the clear information g included in the multi-pulse data f1 manually inputted from the encoder 10a described above. This is a synthesis filter internal data save/clear means for clearing or saving internal data.

FIG. 4 shows an operation flowchart of the decoder of FIG. 3. In FIG. 4, the portion indicated by the dashed dashed line is the operation of the portion corresponding to the multi-pulse audio encoding method of the present invention.

First, multi-pulse data f including the above clear information g,
is input to the decoder 31 of the decoder 30a, and the quantized pulse information decoded by the decoder 31 is dequantized by the dequantization means 32, inputted as pulse data C to the synthesis filter means 34, and decoded in the same manner. The quantized synthesis filter coefficients are inversely digitized by the inverse quantization means 33, and are sent to the synthesis filter means 34 as synthesis filter coefficients d.
(step 521). Furthermore, the decoder synthesis filter internal data is loaded into the synthesis filter means 34 by the synthesis filter internal data loading means 35, and the synthesis filter means 34 synthesizes the speech to produce the speech signal a.
1 (step 523). audio signal a1
The synthesis filter internal data saving 7/clearing means 36a which manually inputs the above information detects the input clear information tlftg and uses this clear information g to clear the decoder synthesis filter internal data stored in a memory means (not shown). Step 524)
If an instruction is given to clear the decoder synthesis filter internal data, the memory means (not shown) is cleared in step S25, and if an instruction is given not to clear the synthesis filter internal data, it is cleared. The decoder synthesis filter internal data is saved in the illustrated storage means (step 526).

[Effects of the Invention] As explained below, according to the multipulse speech encoding method of the present invention, the encoder stores data in the first memory when the power of the synthesized speech is smaller than a predetermined threshold. Clearing the encoder synthesis filter internal data stored in the encoder synthesis filter prevents the effects of errors in the encoder synthesis signal due to transmission bit errors from lingering into subsequent frames in the encoder. In addition, if the clear information sent from the encoder instructs to clear the decoder synthesis filter internal data, the decoder clears the decoder synthesis filter internal data stored in the second memory. Clearing the internal data prevents the effects of errors in the decoder's composite signal due to transmission bit errors from lingering into subsequent frames in the decoder.

Therefore, unlike the conventional multi-pulse audio encoding method, it is possible to prevent the influence of a synthesized signal in which an error has occurred from lingering on subsequent frames.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an encoder in an embodiment of the multi-pulse speech encoding method of the present invention, FIG. 2 is an operation flowchart of the encoder, and FIG. 3 is a block diagram of a decoder in the embodiment. 4 is an operation flowchart of a decoder, FIG. 5 is a block diagram of a conventional encoder, and FIG. 6 is a block diagram of a conventional encoder.
The figure is a block diagram of a conventional decoder. 1-Oa... Encoder, 24... Synthetic voice power detection 3 means, 30a... Decoder, 36a... Synthesis filter internal data save/clear means, g... Clear information.

Claims (1)

[Claims] A first memory that encodes audio into multi-pulse data and outputs it in units of frames, and stores internal data of an encoder synthesis filter for calculating an influence signal for the next frame. an encoder; and a second memory for storing internal data of a decoder synthesis filter for decoding the multi-pulse data frame by frame to reproduce audio and calculating an influence signal for the next frame. In a multi-pulse speech encoding method comprising a decoder, the encoder stores the synthesized speech in a first memory if the power of the synthesized speech is smaller than a predetermined threshold after multi-pulse encoding. clearing the internal data of the encoder synthesis filter, and sending clear information indicating that the internal data of the decoder synthesis filter is to be cleared and multi-pulse data to a decoder; After decoding, if the clear information sent from the encoder instructs to clear the decoder synthesis filter internal data, the decoder synthesis filter stored in the second memory A multi-pulse audio encoding method characterized by clearing internal data.