Chinese phonetic symbol tone identification method based on improved tone core model

Abstract

The invention provides a method capable of improving the average identification rate of four kinds of tones. According to the method capable of improving the average identification rate of the four kinds of tones, an improved tone core model is provided, and an algorithm for extracting an isolated syllable fundamental frequency curve tone core in a self-adaptation mode is designed. The algorithm comprises the steps that outliers sections on the head portion and the tail portion are firstly removed; tone core decision is conducted on the combined subsection result after Viterbi partition and T hypothesis testing, and a tone core is obtained. Three groups of experiments for tone identification with the utilization of different methods are designed; the experiment test A extracts the acoustic feature of the whole syllable to conduct tone identification, the experiment B extracts the acoustic feature of the tone core on the basis of a traditional tone core model, and the experiment C extracts the acoustic feature of the tone core in a self-adaptation mode with the utilization of the improved tone core model to conduct tone identification. The experimental result indicates that the tone extracted by the improved tone core model is used for conducting tone identification, and therefore the average identification rate of the four kinds of tones is improved.

Description

(1) Technical field: [0001] The invention relates to a Chinese speech tone recognition method based on an improved tone kernel model, which belongs to the field of information technology. (2) Background technology: [0002] Tone recognition is an important research content of speech recognition. According to the speech excitation modulation model, the speech signal generation includes glottal excitation and vocal tract modulation. Glottal excitation determines the change of speech prosody and plays an important role in speech emotion recognition. Vocal tract modulation mainly determines the speech content, and each vowel corresponds to a different formant, reflecting different vocal tract shape information. Chinese is a tonal speech, most syllables are composed of initials and finals, syllables composed of the same initials and finals have different meanings and express different emotions depending on the tone. There are 4 main tones in Chinese, including: Yinping, Yangpi...

AI Technical Summary

Problems solved by technology

However, the tone kernel model ignores the outliers in the fundamental frequency curve, and only divides the fundamental frequency curve into the mode of transition section + tone kernel + transition section

In fact, since any syllable segmentation algorithm cannot 100% accurately segment the voiced part and the unvoiced part, there are some outliers in the base frequency of the unvoiced segment at the beginning and end of the base frequency curve. Frequency outliers will lead to errors in the extraction of tone kernels using the traditional three-segment tone kernel model

Images