JPH07123227B2 - Speech coder - Google Patents

Description

TECHNICAL FIELD The present invention relates to a speech coder in a speech analysis system.

(Prior Art) Conventionally, this type of speech coder has performed speech coding using one type of quantization table. FIG. 2 shows a block diagram of a speech coder according to the conventional method. A linear prediction coefficient is calculated from the input speech signal 1 by the LPC analyzer 2, a prediction signal is calculated by the linear predictor 3, a residual with the input signal is calculated, and the residual quantizer 4 inputs it to the residual quantizer. The configuration is such that the encoded value is obtained, a bit stream is formed by the multiplexer 8, transmitted as the output encoded signal 9, and reproduced by the decoder to obtain the output voice signal.

(Problems to be Solved by the Invention) Since the conventional speech encoder described above performs encoding by using only one type of quantization table, the speech signal that does not conform to the quantization table deteriorates in sound quality. There is a drawback that there are many.

(Means for Solving Problems) The speech coder of the present invention includes 2 ^N kinds of quantization tables of linear prediction coefficients and the like, obtains a decoded signal using each table, and obtains an error from the input signal. It comprises means for obtaining power, means for selecting a quantization table so that its error power is minimized, and means for transmitting the selected quantization table to the decoder in N bits. That is, by providing 2 ^N kinds of quantization tables, the number of non-conforming quantization tables due to audio signals is extremely reduced. As a result, accurate quantization is possible and sound quality deterioration is reduced.

(Embodiment) Next, one embodiment of the present invention will be described with reference to FIG.

The speech coder of the illustrated embodiment has a decoder 5 and an error power calculator 6 as compared with the conventional speech coder of FIG. 2, and further includes a linear predictor 3, a residual quantizer 4, The decoder 5 and the error power calculator 6 are increased from one stage to 2 ^N stages, and a quantization table selector 7 is added. Note that N is a natural number. Here, for the determination of the constant N, for example, N = 4
Then, 2 ⁴ = 16 types of tables are provided, and it is possible to satisfy the characteristics of almost all voice signals of men and women of all ages. That is, 16 types of blocks of the linear predictor 3, the residual quantizer 4, the decoder 5 and the error power calculator 6 are realized by a microprocessor capable of high-speed product sum calculation, and the block of the quantization table selector 7 is realized. Is realized by a microprocessor capable of comparison calculation.

As described above, according to the present invention, the quantization table is
By providing 2 ^N types and encoding, there is an effect that the deterioration of the sound quality can be suppressed to a minimum regardless of the type of the audio input signal.

[Brief description of drawings]

FIG. 1 is a block diagram showing a circuit configuration of a speech coder according to an embodiment of the present invention, and FIG. 2 is a block diagram of a speech coder according to a conventional system. 1 ... Input speech signal, 2 ... LPC analyzer, 3 ... Linear predictor, 4 ... Residual quantizer, 5 ... Decoder, 6 ... Error power calculator, 7 ... Quantizer Table selector, 8 ... Multiplexer, 9 ... Output encoded signal.

Claims (1)

[Claims] 1. A speech coder for coding a speech signal or the like with linear prediction coefficients or the like, comprising 2 ^N kinds of quantization tables for quantizing parameters such as linear prediction coefficients (N is a natural number), And a means for obtaining 2 ^N kinds of decoded signals to obtain an error power with respect to the input signal, a means for selecting a quantization table so that the error power is minimized, and a decoder for selecting the kind of the selected quantization table. Means for determining the error power from the input signal and all the 1 to 2 ^N stages of the decoder are constituted by a microprocessor capable of high-speed multiply-accumulate operation, and the quantization table A speech coder, characterized in that the means for selecting is composed of a microprocessor capable of comparison calculation.