JPH07287599A - Voice-coder - Google Patents

Abstract

PURPOSE:To make encoded voice quality excellent when a prediction error signal found by a prediction filter is quantized by linearly predicting and analyzing an input signal. CONSTITUTION:The voice-coder is provided with an impulse response calculator 13 which finds the impulse response of the prediction filter 12 generated from the output of a linear predictive analyzer 11, an impulse response frequency area converter 14 develops the found impulse response in a frequency range, and a formant calculator 15 which finds a formant calculates the position of the formant. Further, the prediction error signal outputted by the prediction filter 12 is developed by a prediction frequency range converter 16 into a frequency range prediction residue signal. The formant part of the prediction error of the output of the prediction filter 12 is quantized by a formant quantizer 17 and other parts are quantized by a quantizer 18.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coder and, more particularly, to a prediction residual quantizer.

[0002]

2. Description of the Related Art As a method for compressing and coding a speech signal, there is a speech coding apparatus using linear prediction analysis.

FIG. 3 shows the configuration of a conventional speech coding apparatus of this type. In FIG. 3, 1 is a linear prediction analyzer, 2 is a prediction filter, and 3 is a prediction residual quantizer.
Further, a is an input voice signal, b is a filter coefficient, and c is a prediction residual signal.

The operation of the above conventional example will be described below. The linear prediction analyzer 1 regards the vocal tract information of the input voice signal a as a filter and calculates the filter coefficient thereof. The prediction filter 2 converts the audio signal into a prediction residual signal c according to the filter coefficient b obtained by the linear prediction analyzer 1. The prediction residual signal c is quantized by the prediction residual quantizer 3 to obtain a quantized prediction residual signal.

In the apparatus of this system, the signal to be transmitted to the decoding unit is the filter coefficient b obtained by the linear prediction analyzer 1,
It is a quantized prediction residual signal quantized by the prediction residual quantizer 3. Then, in the decoding unit, a prediction residual signal is obtained from the quantized prediction residual signal and is passed through a synthesis filter to obtain a decoded speech signal.

[0006]

However, a human voice signal has a characteristic called a formant, and there is a bias between a portion having strong frequency and a portion having weak frequency power. This is one of the characteristics of a speech signal, and in the above-described conventional speech coding apparatus, quantization is performed equally for all frequency bands, and since quantization that does not consider this characteristic is not performed, the formant There is a problem that the portion cannot be quantized faithfully and good sound quality cannot be obtained.

The present invention solves the above-mentioned conventional problems, and provides a speech coding apparatus capable of reproducing excellent sound quality by quantizing a prediction residual signal in consideration of formant information. The purpose is to

[0008]

In order to achieve the above object, a speech coder according to the present invention is a speech coder for performing linear prediction analysis on input speech and quantizing a prediction residual obtained by a prediction filter. , A means for specifying a formant part of a voice signal and quantizing the formant part independently of other parts is provided.

[0009]

According to the present invention, with the above configuration, a good decoded speech can be obtained by effectively quantizing the waveform characteristic of the speech signal in consideration of the formant which is one of the characteristics of the human speech signal. realizable.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a speech coder according to an embodiment of the present invention. In FIG. 1, 11 is a linear prediction analyzer that calculates the filter coefficient of the input speech signal, and 12 is a prediction filter that converts the obtained filter coefficient into a prediction residual signal. Reference numeral 13 is an impulse response calculator that calculates the impulse response of the prediction filter 12. Reference numeral 14 is an impulse response frequency domain converter that expands the obtained impulse response into a frequency domain impulse response. A formant calculator 15 calculates a formant part from the frequency domain impulse response. 16
Is a prediction residual frequency domain converter that expands the prediction residual signal obtained by the prediction filter 12 into a frequency domain prediction residual signal. A formant quantizer 17 quantizes the frequency domain prediction residual signal corresponding to the formant part based on the information from the formant calculator 15. A quantizer 18 quantizes the frequency domain prediction residual signal other than the formant part. a is an input speech signal, b is a filter coefficient, c is a prediction residual signal, and d is an impulse response of the prediction filter 12. e is a frequency domain impulse response in which the impulse response d is expanded in the frequency domain, f is a formant partial deterministic signal, and g is a frequency domain prediction residual signal in which the prediction residual signal c is expanded in the frequency domain.

The operation of the speech coding apparatus configured as above will be described below. First, the linear prediction analyzer 11 calculates the filter coefficient b of the prediction filter 12 from the input audio signal a. The prediction filter 12 uses the filter coefficient b to convert the input speech signal a into a prediction residual signal c. Further, the impulse response calculator 13 calculates the impulse response d of the prediction filter 12 from the filter coefficient b. The prediction residual signal c and the impulse response d obtained above are signals on the time axis.

Next, the prediction residual frequency domain converter 16 transforms the prediction residual signal c into a frequency domain prediction residual signal g. Similarly, the impulse response frequency domain converter 14 develops a frequency impulse response e which is a signal on the frequency axis. The frequency domain impulse response e is the input speech signal a
The power of each frequency is shown. From this power value, the formant, which is one of the characteristics of the input audio signal a, can be estimated as shown in FIG. Also, from the frequency domain impulse response e, the formant calculator 15 determines the frequency domain prediction residual signal g
A formant part confirmation signal f indicating which of them corresponds to the formant part is calculated.

Next, the frequency domain prediction residual signal g corresponding to the formant part is determined by the formant part determination signal f and quantized by the formant quantizer 17. The frequency domain prediction residual signal g that does not correspond to the formant part is quantized by the quantizer 18.

The signal sent to the decoder is the linear prediction analyzer 1.
1 is the filter coefficient b obtained in step 1, the formant partially quantized residual signal which is the output of the formant quantizer 17, and the quantized residual signal which is the output of the quantizer 18. Then, the decoding unit decodes the prediction residual signal from the formant partially quantized residual signal and the quantized residual signal, and passes it through a synthesis filter to obtain a decoded speech signal.

As described above, according to the above embodiment, the formant portion of the voice signal is quantized independently of the other portions, so that the waveform characteristics of the voice signal can be maintained even in the decoded voice, which is excellent. It is possible to play the decoded audio.

[0016]

As described above, the present invention has means for specifying the formant portion of a voice signal and quantizing the formant portion independently of the other portions. Therefore, the waveform characteristic of the voice signal is decoded. This has the effect that even voice can be maintained, and good decoded voice can be reproduced.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a speech coding apparatus according to an embodiment of the present invention.

FIG. 2 is a characteristic diagram showing a signal spectrum calculated from an impulse response of a filter coefficient.

FIG. 3 is a block diagram showing a schematic configuration of a speech encoding apparatus in a conventional example.

[Explanation of symbols]

 11 Linear Prediction Analyzer 12 Prediction Filter 13 Impulse Response Calculator 14 Impulse Response Frequency Domain Transformer 15 Formant Calculator 16 Prediction Residual Frequency Domain Transformer 17 Formant Quantizer 18 Quantizer

Claims (2)

[Claims] 1. A speech coding apparatus for linearly predicting an input speech and quantizing a prediction residual obtained by a prediction filter, wherein a formant portion of a speech signal is specified and quantized independently of other portions. A speech encoding apparatus having means for performing. 2. A linear prediction analyzer for calculating a filter coefficient of an input speech signal, a prediction filter for converting the obtained filter coefficient into a prediction residual signal, and an impulse response calculation for calculating an impulse response of the filter coefficient. A vessel,
An impulse response frequency domain converter that expands the impulse response into a frequency domain impulse response, a formant calculator that calculates a formant part from the frequency domain impulse response, and a prediction residual signal obtained by a prediction filter from a frequency domain prediction residual. A predictive residual frequency domain converter expanded into a signal, a formant quantizer that quantizes a portion corresponding to the formant part of the frequency domain residual based on information from the formant calculator, and a frequency other than the formant part. A speech coding apparatus comprising: a quantizer for quantizing a region prediction residual signal.