JPS6222489B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention encodes the difference between an input signal and a predicted signal, in which sample values are correlated, such as an audio signal, and locally decodes the difference, and generates the predicted signal from the decoded signal. The present invention relates to a predictive encoding device.

FIG. 1 shows the configuration of a conventional speech predictive coding device. The difference between the audio signal input from the input terminal 1 and the prediction signal from the local decoder 2 is taken by a difference circuit 6, and the prediction error signal from the difference circuit 6 is encoded by a quantizer 7 and sent to an output terminal 8. and is also supplied to the local decoder 2 and decoded. A predicted signal is created from this decoded signal. That is, it is determined whether the input audio signal is voiced or unvoiced for each frame of a certain length, and for voiced sounds, prediction is performed from the decoded signal one pitch earlier using the variable prediction circuit 3 in the local decoder 2. For unvoiced sounds, a switch 4 is used to perform fixed prediction using a fixed prediction circuit 5 from adjacent decoded signals. These predicted signals from the switch 4 are supplied to the difference circuit 6 and also to the sum circuit 9 in the local decoder 2, where they are added to the code at the terminal 8 and decoded, and the decoded signal is interlocked with the switch 4. The signal is supplied to the corresponding prediction circuits 3 and 5 through the switch 11.

The variable prediction circuit 3 stores the decoded signal,
The autocorrelation between a certain decoded signal and a plurality of decoded signals preceding it by the pitch period (with distribution width) of the input audio signal was taken, and the previous decoded signal with the highest correlation was used as the predicted signal. Note that the autocorrelation function v〓 (τ: integer) of the audio waveform is the average value of the product of the signal x _t and the signal x _t −τ shifted by τ samples v〓=<x _t・x _t −τ>(<x> is the average value of x), and the autocorrelation coefficient p〓 is p〓=v〓/v ₀
is defined in Note that autocorrelation is a conventional method (function) that is widely used in various statistical processing such as audio signal processing, and is also used in magazines such as Nikkei Electronics (published February 12, 1973, p58-75). Published. Further, if a correlation larger than a predetermined value cannot be obtained when the correlation is calculated, the switches 4 and 11 are switched to the fixed prediction circuit 5 side. Therefore, the voiced/unvoiced discrimination signal, pitch, and frame synchronization signal must be transmitted for each frame in addition to the output code of the quantizer 7 in order to be decoded on the receiving side. The operation was complicated.

In order to eliminate this drawback, the present invention is characterized in that the correlation function of the decoded signal in the local decoder is calculated for each sample input, and the decoded signal for which the correlation function takes the maximum value is used as the predicted signal. This eliminates the need to transmit frame synchronization signals, pitch, and discrimination signals, reduces the amount of transmitted information, simplifies decoding operations on the receiving side, and allows optimal prediction to be performed by always using samples with the highest correlation. be.

For example, when an audio signal is input from the input terminal 1 as shown in FIG. 2 with the same reference numerals attached to the parts corresponding to those in FIG. An autocorrelation function is obtained for the decoded signals decoded and stored within the period, and the decoded signal with the maximum correlation is supplied to the difference circuit 6 as a prediction signal. The difference between the input signal and its predicted signal, ie, the prediction error signal, is encoded and transmitted by the quantizer 7. At the same time, this prediction error signal is added to the prediction signal by an adder circuit 9 and stored in a prediction circuit 12.

The prediction circuit 12 detects the maximum autocorrelation in the same way as the conventional variable prediction circuit 3 shown in FIG. The correlation with the decoded signal is taken, and the prediction circuit 12 takes the correlation with the decoded signal 1 to 120 samples before and detects the maximum one. When sampling an audio signal at 8 KHz, the prediction circuit 12 only needs to store decoded signals for about 240 samples.

To decode the signal encoded in this manner on the receiving side, a circuit similar to the local decoder 2 can be used to reproduce the audio signal.

As described above, the speech predictive coding device according to the present invention can always perform optimal prediction, and compared to conventional adaptive differential PCM using adjacent signals, when configured with the same quantization circuit, the prediction is about 3 dB.
It is possible to improve SNR. Further, since there is no need to transmit a frame synchronization signal, a voiced/unvoiced discrimination signal, and a pitch period, it is possible to reduce the amount of information transmitted.

As explained above, the device according to the present invention uses a decoded signal for each input sample to make predictions from the sample with the maximum correlation. Therefore, optimal prediction is always possible, and additional information such as frame synchronization signals and pitches are added. The advantage is that there is no need to transmit.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a conventional speech predictive coding device, and FIG. 2 is a block diagram showing an embodiment of the speech predictive coding device according to the present invention. 1: Audio input terminal, 2: Local decoder, 7: Quantizer, 8: Output terminal, 12: Prediction circuit.

Claims (1)

[Claims] 1. Detecting the difference between a sampled input signal and a predicted signal for each sample, quantizing and encoding the prediction error, and decoding the encoded signal for each sample by a decoding means. , a predictive coding device that generates the predicted signal from the decoded signal, comprising: means for storing the decoded signal; and a decoded signal decoded by the decoding means and a decoded signal of a sample approximately 1 to 120 samples before the decoded signal. A predictive encoding device comprising means for determining an autocorrelation coefficient between the two samples and outputting the decoded signal of the sample about 1 to 120 times before, in which the correlation coefficient is maximum, as the prediction signal.