Voice HNR automatic analytical method

Abstract

The invention provides an automatic voice harmonic-to-noise ratio analysis method, which comprises: 1) an effective phonetic segment for harmonic-to-noise ratio analysis is segmented from the recording; 2) the phonetic segment is subject to filtering processing based on an auditory model, and then a two-dimensional energy correlation coefficient of a time domain and a frequency domain in various filter channels in the auditory model is calculated; and 3) the threshold value of the correlation coefficient is preset, a coordinate point of the time domain and the frequency domain corresponding to the correlation coefficient has a harmonic component when the correlation coefficient obtained in step 2) is larger than the threshold, or else the coordinate point of the time domain and the frequency domain corresponding to the correlation coefficient has a noise component, and finally the ratio of the harmonic component and the noise component is calculated to obtain the harmonic-to-noise ratio. The method uses the correlation between the time domain characterized by an autocorrelogram and a cochlear spectral domain channel to judge the harmonic component, is not affected by the detection position of a fundamental frequency, and can detect the harmonic component more accurately and more robustly. As the cochlear spectrum is used, the method is more matched with the actual hearing of human ears.

Description

technical field [0001] The invention belongs to the technical field of speech signal processing, in particular, the invention relates to a harmonic-to-noise ratio analysis method in automatic voice evaluation. Background technique [0002] Harmonic to Noise Ration (HNR) is the main indicator for analyzing and evaluating long vowels. The traditional calculation method of harmonic-to-noise ratio is to use the autocorrelation of the periodic signal in the time domain to estimate the period of the fundamental frequency of the signal. Then, according to the position of the fundamental frequency cycle, the part with strong correlation near the fundamental frequency cycle is regarded as the harmonic component, and the part with weak or irrelevant correlation is regarded as noise to calculate the harmonic noise ratio. There are some defects in this method: 1. For some voice samples with more serious pathological changes or hoarse voice samples, the estimation of the fundamental fre...

AI Technical Summary

Problems solved by technology

There are some defects in this method: 1. For some voice samples with more serious pathological changes or hoarse voice samples, the estimation of the fundamental frequency period is prone to deviation, and sometimes it is even difficult to estimate the period of the sample, so that It is impossible to calculate the effective harmonic-to-noise ratio value; 2. These harmonic calculations are carried out in the ordinary time-frequency domain, which is very different from the real human ear perception, so in practical applications, the final calculation results Unreasonable match with subjective assessment scores by voice experts

This method adds manual intervention, which is not only time-consuming and laborious, but also has a subjective impact on the results, making it difficult to reproduce the calculation process

Images