A Nonlinear Quantization Method for Speech Linear Prediction Model

Abstract

The invention discloses a nonlinear quantization method of a voice linear prediction model. The method comprises the following steps: line spectral frequency parameter transformation step: transforming line spectral frequency parameters of a voice linear coding prediction model into a line spectrum frequency parameter difference value through linear transformation; nonlinearity decorrelation step: decorrelating the line spectrum frequency parameter difference value through nonlinear transformation by adopting a structured method according to a statistical property of the line spectrum frequency parameter difference value; marginal probability distribution calculation step: calculating marginal probability density distribution by utilizing the statistical property of the decorrelated line spectrum frequency parameter difference value; and scale quantizer design step: designing an optimum scale quantizer based on probability density function according to the obtained marginal probability density distribution function. According to the nonlinear quantization method of the voice linear prediction model, the defects that in the prior art, the time complexity is high, the using effect is poor, and the like can be overcome, so that the advantages of low time complexity and good using effect can be realized.

Description

technical field [0001] The invention relates to the technical field of speech coding, in particular to a nonlinear quantization method of a speech linear prediction model. Background technique [0002] With the in-depth development of Internet technology, voice communication technology has made great progress, and the transmitted voice signal has evolved from narrowband signal propagation to broadband signal propagation. With the continuous development and promotion of multimedia applications, people have higher and higher requirements for voice transmission quality and real-time performance in voice communication technology. Therefore, there is an urgent social need to study efficient and reliable voice communication algorithms. [0003] The primary problem to be solved in voice communication is the coding of voice. After decades of development, speech coding technology can be roughly divided into three ways: waveform coding technology, coding technology based on parameter...

AI Technical Summary

Problems solved by technology

[0005] However, the traditional optimization quantification method based on the probability density distribution is mainly based on the Gaussian statistical model, which cannot describe the bounded and ordered characteristics of the line spectrum frequency parameters well.

In response to this problem, some scholars proposed a bounded Gaussian mixture model to describe the above characteristics of the line spectrum frequency parameters. Although some improvements in quantization performance have been achieved, it has brought about a negative impact of too high time complexity. The overall effect unsatisfactory

[0006] In the process of realizing the present invention, the inventor found that there are at least defects such as high time complexity and poor use effect in the prior art

Images