JPH0650822B2 - Encoder / decoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION "Industrial field of application" The present invention relates to an encoder for separating, for example, a voice signal into its parameter information and time-series information, and then multiplexing and transmitting them on a time axis and its encoder. It relates to a decoder.

"Prior art" Conventionally, as a voice encoding method for digitizing and transmitting voice, adaptive prediction is performed by predictively encoding a voice signal by adaptively changing a coding characteristic such as a quantization width according to its property. There is an encoding method. This coding method is a method of coding and transmitting only the difference between the predicted value linearly predicted from the past several sample values of the input signal and the current sample value, that is, the prediction residual. There are various possible forms of this system depending on the configurations of the predictor and the quantizer. For a non-stationary input signal such as a speech signal, the signal-to-quantization noise ratio (S / S) is changed by changing the characteristic parameters such as the prediction coefficient and the quantization width so as to match the statistical properties of the signal. N
It has been proposed to improve the ratio). This feature parameter is not sent directly to the decoder. For example, there is a method of increasing the quantization width when the sequence of maximum values is detected by looking at the received prediction residual information sequence. However, by using the conventional adaptive predictive coding method, the quality required for the telephone is achieved, and the processing ends for each sample value, that is, in order to perform the sequential processing, information of 32 kb / s or more is required. Speed is needed.

Therefore, at an information speed of 16 kb / s or less, the S
In order to obtain the / N ratio, the input audio signal sequence is divided into certain frames (windows) at regular time intervals, and the entire signal consisting of a plurality of sample values in the frame is analyzed to The characteristic parameters (for example, prediction coefficients) indicating the characteristics of the entire signal are extracted, parameter information that is a set of these characteristic parameters is obtained, and the information is notified to the decoder, and predictive coding is performed based on this parameter information. Then, it is necessary to generate a prediction residual signal for each sample, obtain time series information that is an aggregate thereof, separate each, and quantize and transmit each. It should be noted that the frame here does not mean a "constant number of bits" used in time division multiplex transmission, but has almost the same meaning as a window in the field of digital signal processing (DSP), and is a speech coding technique. It is a term used in the field. Hereinafter, in order to avoid confusion, a frame used in time division multiplex transmission will be referred to as a “transmission frame”. In the method of obtaining the time series information after obtaining the parameter information by dividing the frame (window) unit, it is necessary to construct and analyze the frame, and therefore, the processing is completely true for each sample in both coding and decoding. It cannot be processed.

As an adaptive predictive coding method in such a low bit region, Japanese Patent Application No. 59-25630 proposes an adaptive predictive coding method of a format in which a frame is constructed and parameter information and time series information are transmitted to a decoder. Has been done. In this system, the encoder side separates into two types of signals, parameter information (pitch prediction coefficient, pitch period, pitch position, partial interval residual power, total residual power) and time series information (prediction residual), Encode and send each to the transmission path. On the other hand, the decoder side decodes the above parameter information, sets various characteristic parameters of the decoder based on the parameter information, and then sequentially decodes by adding the time series information, and resynthesizes the original time series audio signal. .

A conventional example in which such a method of constructing a frame and extracting a characteristic parameter is performed in real time by a digital signal processor (this is a term for storage and batch processing) is shown in FIG. 6 and subsequent drawings. Let's explain with reference to. That is, in the encoder, as shown in FIG.
Only the frame length is temporarily stored in the input buffer processing unit 2. Next, with respect to the signal stored in the input buffer processing unit 2, the signal analysis unit 3 in the transmission signal processing unit 100 extracts the characteristic parameter of the input signal in the frame, and the parameter encoding unit 4 encodes it. Further, the adaptive predictive coding unit 5 adaptively predictively codes the original signal stored in the input buffer processing unit 2 based on the parameters extracted by the signal analysis unit 3, and codes the predictive residual signal. This signal is called 0-series information. The parameter information coded by the parameter coding unit 4 in the transmission signal processing unit 100 and the time-series signal coded by the adaptive prediction coding unit 5 are multiplexed by the transmission buffer processing unit 6 and one frame worth. When the data is stored and one frame is stored, it is sent to the transmission path through the transmission buffer processing unit 6 and the output terminal 7. The transmission buffer processing unit 6 has a so-called double buffer structure.

On the other hand, on the decoder side, as shown in FIG. 7, the code sequence input from the transmission line through the input terminal 8 is temporarily stored in the reception buffer processing unit 9 for the encoded frame length. This part also has a double buffer structure. The code sequence stored in the reception buffer processing unit 9 is the reception signal processing unit 200.
The parameter / time-series information separation unit 10 therein separates the parameter information and the time-series information, and the parameter information among them is decoded by the decoding unit 11, and the characteristic parameters of the decoding unit 12 are set. Parameter / time series information separation unit 1
The time series information extracted by 0 is repeatedly subjected to adaptive prediction decoding for each sample in the frame in the decoding unit 12, and the decoded signal is temporarily stored in the output buffer processing unit 13.
The decoded signal stored in the output buffer processing unit 13 is output to the output terminal 14 for each sampling period, and the reproduction of the audio signal ends.

This conventional encoding / decoding process will be described again according to the actual hardware configuration. In the encoder shown in FIG. 8, the digital signal sampled value input from the input terminal 1 is synchronized with the reference clock of the reference clock generator 15 for supplying the sampling interval, the frame period, the transmission bit rate, etc. The coded frame length is stored in the input buffer 17 according to the input command sent from the input buffer control unit 16. At this time, the transmission signal processing unit 100 sets 1
The input signal stored in the input buffer 18 is read in the immediately preceding frame period, the characteristic parameters are extracted, encoded and adaptively predicted, and separated into parameter information and time series information. At this time, the input buffer switching unit 19
It is assumed that the input buffer switching unit 20 is connected to the 9a (input buffer 17) side and the input buffer switching unit 20 is connected to the 20b (input buffer 18) side. This structure is called a so-called double buffer. The separated code sequence (the parameter information and the time-series information are collectively referred to as a code sequence) is sent out and stored in the transmission buffer 22. At this time, the code sequence of one frame before stored in the output buffer 23 in the frame period of one frame before is transmitted from the transmission buffer controller 24 in synchronization with the reference clock from the reference clock generator 15. Is transmitted to the transmission line through the terminal 7 bit by bit. At this time, the transmission buffer switching unit 21 is connected to the 21a (transmission buffer 22) side, and the transmission buffer switching unit 25 is connected to the 25b (transmission buffer 23) side. Input buffer switching unit 1
9, 20 and transmission buffer switching units 21, 25
The input buffer control unit 16 and the transmission buffer control unit 24 generate switching signals in response to the clock signal from the reference clock generation unit 15, and the opposite terminal, that is, 19a to 19b side, 20b side to 20a side in frame length units. 21a side to 21b independently of this.
Side, it is simultaneously switched from the 25b side to the 25a side.
Here, the input buffer control unit 16 sends to the transmission signal processing unit 100 a signal indicating that one frame of input signals to be processed has been accumulated at the same time as switching. The transmission signal processing unit 100 sends a transmission signal processing end command to the transmission buffer control unit 24 after the processing of the signal for one frame to be processed is completed, and the transmission buffer control unit 24 receives the instruction. The code sequence is transmitted to the transmission path.

FIG. 9 shows a conventional example in which a signal is decoded from a code sequence which is the result of coding in this way. The code sequence input from the terminal 8 of the transmission line is stored in the reception buffer 30 in synchronization with the reference clock of the reference clock generation unit 26 in the reception buffer 30 according to the reception command transmitted from the reception buffer control unit 27.

[Problems to be Solved by the Invention] The delay times caused by the conventional encoder / decoder are shown in FIG. In other words, the main delay time in the encoder is delay: sample value acquisition time of one encoded frame generated in the double buffer (processing (i) in FIG. 6), delay: one frame for extraction and encoding of parameter information. Time required for processing in units (processing (ii) in FIG. 6), delay: processing time for processing adaptive predictive coding in units of one frame (processing (ii in FIG. 6
i)), delay: transfer output processing time (processing (iv) in FIG. 6) to the transmission buffer generated in the double buffer. On the other hand, the main delay time in the encoder is the delay:
Code sequence acquisition time (processing (i) in FIG. 7) of one encoded frame generated in the double buffer, delay: time required for separation and decoding of parameter information (processing (ii) in FIG. 7), delay : Adaptive prediction decoding processing time (processing (iii) in FIG. 7), delay: transfer to output buffer and output processing time (processing (iv) in FIG. 7).

Therefore, the total delay time until the input signal A to the encoder at the time t = 0 is output as the decoded signal A'from the decompressor is the sum of the above-mentioned delays.

For example, the coded frame length is 20 ms, and one instruction execution time of the digital signal processor is 250 ns.
Assuming that, the delay and the delay each require 20 ms, which is the same as the coded frame length. Also delay,
The total delay time is about 50 ms because the sum of the delay time and the delay time requires about 5 ms each.

If you have a conversation using such an encoder / decoder,
There is a drawback that the conversation becomes unnatural due to the echo (wraparound of the handset) and the delay in response due to the total delay. Further, in the encoder and the decoder, the input buffer, the transmission buffer, the reception buffer, and the output buffer each have a so-called double buffer configuration in which two or more memories having the same capacity are required, and the hardware scale is inevitably large.

An object of the present invention is to have a relatively low information speed of 8 to 16 kb
In the area of about / s, the present invention proposes an encoder / decoder capable of real-time processing with a short delay time and a small hardware scale in a frame processing type high-efficiency speech coded signal.

"Means for Solving the Problem" The present invention limits only the parameter information that is temporarily stored in the reception buffer of the decoder, and the time-series information is based on the above parameter information without being stored in the reception buffer. It is characterized by performing sequential processing for each sample.

[Operation] In the present invention, time series information having a long signal length is processed immediately after reception without being temporarily stored in the reception buffer.
The processing delay corresponding to the time series information can be eliminated, and the delay time required by the encoder / decoder can be shortened. In this respect, it is remarkably different from the conventional method which requires a delay of one frame unit for each processing. Further, the hardware scale can be reduced by configuring the reception buffer processing unit 47 and the output buffer processing unit 51 in the decoder with a single buffer configuration instead of the double buffer configuration.

[Embodiment] An embodiment of the encoder / decoder according to the present invention will be described below. FIG. 1 shows an embodiment of the encoder. The sampled input signal from the input terminal 1 which has been converted into a digital signal is temporarily stored in the input buffer processing section 2 for the encoded frame length. Next, the signal of the input buffer processing unit 2 is extracted by the signal analysis unit 3 from the characteristic parameters of the input signal, and the parameter to be transmitted to the decoder is parameter-encoded by the parameter encoding unit 4. The processing up to this point is the same as the processing from the terminal 1 to the parameter coding unit 4 in FIG. Next, the adaptive prediction coding unit 5 performs adaptive prediction coding based on the parameters extracted by the signal analysis unit 3, and codes the prediction residual signal (time series information). Further, in the transmission signal processing unit 300, the parameter information is arranged at the head of the code sequence coded by the parameter coding unit 4 and the adaptive prediction coding unit 5, and multiplexed by the bit number of the coding length, and the transmission buffer processing unit. Send to 6. The transmission buffer processing unit 46 sends the contents of the transmission buffer to the transmission line through the terminal 7 in synchronization with the information rate of the transmission line.
On the other hand, on the decoder side, as shown in FIG. 2, only the parameter information is temporarily stored in the reception buffer by the reception buffer processing unit 47 in the code sequence input from the transmission line through the terminal 8. Next, the received signal processor 4 receives the above parameter information.
The parameter decoding unit 48 in 00 performs parameter decoding. That is, the characteristic parameter is set in the decoder. During this time, the reception buffer processing unit 47 sequentially stores the time-series information located after the parameter information in the reception buffer, and the adaptive prediction decoding unit 49 in the reception signal processing unit 400 first decodes by the parameter decoding unit 48. Based on the converted parameter information, the time series information of the reception buffer is sequentially, that is, sequentially, for example, adaptive prediction decoding is performed for each sample. The decoded signal is sequentially stored in the output buffer processing unit 51 and is output at each sampling cycle. The adaptive prediction decoding process for each sample and the output are continued by the number of samples in the frame. At this time, the time series information is received by the buffer processing unit 47.
Since it is stored in the reception buffer at 1, the storage delay time is required only for the time determined by the maximum number of information bits assigned in the time axis direction of the residual output and the time determined by the total number of information bits assigned to the parameter information.

Next, the encoding / decoding process of the present invention will be described according to the actual hardware configuration. As shown in FIG. 3, in the encoder, the sampled value input as a digital signal from the input terminal 1 is synchronized with the input command sent from the input buffer control section 16 in synchronization with the basic clock from the basic clock generation section 15. Then, only the encoded frame length is stored in the input buffer 17. During this period, the transmission signal processing unit 300 reads the input signal one coded frame before stored in the input buffer 18 by the above-described method, performs parameter analysis, parameter coding, and adaptive prediction coding to obtain the parameter information. Separated into time series information. At this time, the input buffer switching unit 19 is connected to the side 19a and the input buffer switching unit 20 is connected to the side 20b. The separated code sequence is rearranged in the transmission processing unit 300 so as to transmit the parameter information first, and then is transmitted to the transmission buffer 22 and stored in the same buffer. During this time 1
The code sequence processed by the transmission signal processing unit 300 before the frame and stored in the transmission buffer 23 is synchronized with the reference clock from the reference clock generating unit 15 by a transmission command sent from the transmission buffer control unit 24. It is sent to the transmission line every one bit. At this time, the transmission buffer switching unit 21 is on the 21a side, and the transmission buffer switching unit 25 is 25b.
It is assumed that they are connected to each side. Input buffer switching units 19 and 20 and transmission buffer switching units 21 and 25
The input clock control unit 16 and the transmission buffer control unit 24 convert the clock signal from the reference clock generation unit 15 into a switching synchronization signal, and the terminal on the opposite side, that is, from the 19a side to the 19b side, to the 20b side in frame length units. From 20a
Side is switched to the side simultaneously, and independently of this, it is switched from the side 21a to the side 21b and from the side 25b to the side 25a. Further, the input buffer control unit sends a read command to the transmission signal processing unit at the same time as switching, and the transmission buffer control unit 24 sends a transmission command upon receiving a transmission signal processing end signal from the transmission signal processing unit 300. Thus, the code sequence is transmitted to the transmission path.

FIG. 4 shows an example of decoding a signal from a code sequence encoded in this way. The input code sequence framed from the terminal 8 is sequentially stored in the reception buffer 30. Here, in the reception buffer control unit 27, the reference clock generation unit 2
In synchronization with the clock signal from 6, the number of information bits pre-allocated to the parameter information is counted, and when the total number of information bits of the parameter information is reached, the received signal processing unit is read by the read command from the reception buffer control unit 27. 400 reads the code sequence stored in the reception buffer 30, and the reception signal processing unit 400 performs parameter decoding from the code sequence of the parameter information by the method described above. After that, after the parameter information is decoded by the reception signal processing unit 400, all parameters for predictive decoding the time series information are set. Next, the time-series information in the frame is stored in the reception buffer 30 in the same direction as the storage of the parameter information. Next, the reception buffer control unit 27 counts the number of information bits assigned to each time series information in synchronization with the clock signal from the reference clock generation unit 26, and when the number of assigned information bits is reached, the reception buffer control unit 27 The reception end signal sent from 27 to the reception signal processing unit 400 causes the reception signal processing unit 400 to read the code sequence stored in the reception buffer 30, and the reception signal processing unit 400 reads the code sequence based on the previously decoded parameter information. Adaptive prediction decoding is performed for each sample. When one sample is decoded, one sample of the decoded signal is stored in the output buffer 35. When one sample is decoded in the reception signal processing unit 400, the output buffer control unit 33 receives the processing end signal transmitted from the reception signal processing unit 400, and the reference clock generation unit 2
The output command is sent to the output buffer control unit 33 after being delayed by a few bits in the delay time setting unit 61 in synchronization with the clock signal from 6. The output buffer unit 35 sequentially starts the output of the decoded signal for each sample in synchronization with the output command. The decoding process of the series of time-series information is sequentially performed for each sample. That is, it is not put in the double buffer as in the prior art. The delay time setting unit 61 delays the output command because the number of information bits assigned to each time-series information varies, so that the time corresponding to the maximum number of information bits among them must be guaranteed. This is because the continuity of the output signal cannot be maintained.

The delay time caused in the encoder / decoder according to the present invention is as shown in FIG. The main delay time in the encoder is delay: 1 sampled sample acquisition time of encoded frame (process (i) in FIG. 1), delay: time required for extraction and encoding of parameter information (process in FIG. 1) (ii)), delay:
Adaptive prediction coding processing time (processing (iii) in FIG. 1) and delay: transfer output processing time to the output buffer (processing (iv) in FIG. 1). On the other hand, the main delay time in the decoder is the delay: the code sequence acquisition time of the parameter information at the head within the length of one encoded frame length (the processing in FIG. 2).
(i)), delay: time required for decoding parameter information (processing (ii) in FIG. 2), delay: acquisition time of a code sequence determined by the maximum number of allocated bits of each sample of time series information, and 1-sample adaptive prediction decoding processing time (processing (iv) in FIG. 2), delay: transfer to output buffer / output processing time (processing (iv) in FIG. 2). Therefore, time t =
The input signal B to the encoder at 0 is from the decoder to the decoded signal B '
The total delay time until being output as is the sum of the above delay times. If the coded frame length and the one-instruction execution time of the digital signal processor are made equal to the conditions described in the prior art and the frame length of the parameter information is assumed to be 4 ms, the total delay time is about 34 ms.
Here, the reference clock generation unit 15 and the reference clock generation unit 26 may be the same. Further, in the transmission buffer processing unit 46 in FIG.
From the transmission buffer 22 or 23 until the transmission signal processing unit 300 completes the encoding process of the parameter information and the time-series information. Transmission buffer 2
2 or 23, the time-series information sent to the transmission path at the time of being stored in 2 or 23 is sequentially subjected to adaptive prediction coding one sample at a time in the transmission signal processing unit 300, and the transmission buffer 22 or 23 is transmitted.
Alternatively, the data may be stored in the network and transmitted to the transmission line. In this case, the transmission signal processing unit 300 to the transmission buffer control unit 24
The transmission command sent to is delayed by the time length obtained by adding the number of bits of the parameter information and the maximum number of bits assigned to the time series information, and the output processing is started. Furthermore, the present invention is a feasible encoder / decoder as long as the adaptive predictive encoding method used is a method that can be separated into parameter information and time series information.

[Advantages of the Invention] As described above, the encoder / decoder according to the present invention first transmits the parameter information to be transmitted to the transmission path by the encoder, and then transmits the time-series information accompanying the parameter information. To do. In the decoder, since the parameter information is first decoded, the time series information can be sequentially processed without being stored in the double buffer. Therefore, the delay time is, for example, 16 m, as compared with the conventional technique in which all the processing is conventionally performed in units of encoded frames.
It can be shortened by about s. Further, since it is not necessary to make the reception buffer and the output buffer of the decoder double, it is possible to reduce the hardware scale to one.

[Brief description of drawings]

FIG. 1 is a block diagram showing the principle of the encoder side of the encoder / decoder of the present invention, FIG. 2 is a block diagram showing the principle of the decoder side thereof, and FIG. 3 is a concrete example of the encoder of the present invention. Block diagram shown in FIG. 4, FIG. 4 is a block diagram showing a concrete example of a decoder in the present invention, FIG. 5 is a diagram showing a delay from the start of encoding to the end of decoding in the present invention, and FIG. 6 is a conventional encoder / decoder. Block diagram showing the principle of the encoder side of the encoder, No. 7
FIG. 8 is a block diagram showing the principle of the decoder side, FIG. 8 is a block diagram showing a concrete structure of a conventional encoder, FIG. 9 is a block diagram showing a concrete structure of a conventional decoder, and FIG. It is a figure which shows the delay from the conventional encoding start to the decoding end. 1: input terminal, 2: input buffer processing unit, 3: signal analysis unit, 4: parameter coding unit, 5: adaptive prediction coding unit,
6: transmission buffer processing unit, 7: output terminal, 100: transmission signal processing unit, 9: reception buffer processing unit, 10: parameter / time series information separation unit, 11: parameter decoding unit, 1
2: Adaptive prediction decoding unit, 13: Output buffer processing unit, 1
4: output terminal, 200: reception signal processing unit, 19, 20:
Input buffer switching unit, 17, 18: input buffer, 2
1, 25: Transmission buffer switching unit, 22, 23: Transmission buffer, 15, 26: Reference clock generation unit, 16: Input buffer control unit, 24: Transmission buffer control unit, 29, 3
2: reception buffer switching unit, 30, 31: reception buffer,
27: reception buffer control unit, 36: output buffer control unit, 34, 37: output buffer switching unit, 35, 36: output buffer, 300: transmission signal processing unit, 400: reception signal processing unit.

Claims (1)

[Claims] 1. A frame unit storage means for sampling an input signal at fixed time intervals and storing a fixed number of the sampled signals, and a signal analysis means for analyzing the stored signal and extracting a characteristic parameter. And a parameter coding means for coding the characteristic parameters to generate parameter information, and an adaptive prediction for adaptively predicting the signals stored in the frame unit storage means based on the characteristic parameters to generate time series information. Of the encoding means, the generated parameter information and the time series information, the parameter information is arranged to be transmitted first, and the transmission signal is generated so that the time series information is arranged to be transmitted thereafter. An encoder having a transmission buffer means for transmitting, and receiving the transmitted signal, separating and outputting the parameter information and the time series information. Receiving buffer means, a parameter decoding means for decoding the parameter information and outputting the characteristic parameter, and receiving the characteristic parameter, and based on this, the time series information received from the receiving buffer means is sequentially A code / decoder comprising a decoding means for performing adaptive prediction decoding, and a decoder having an output buffer for storing the decoded signal and outputting it at each sampling period.