论文部分内容阅读
语谱图在语音分析方面有着广泛的应用.音素的自动分割是语音识别过程中的一个基本阶段,它将把语音句子按音素特征进行分割.本文提出了一个音素自动分割的方法;使用了两个表示语谱图密度变化的形变函数,以及自适应阈值技术来定位每个音素段的边缘.这个方法在计算机上具体实现后.我们对取于一个语谱图数据库的一组实验数据,用本文所介绍的自动分割方法划分音素,将所得结果与由一语音学家分割的结果进行比较,得到的识别率高于93%.这个方法作为语音识别系统的一部分,已经在一个语音分析系统中使用.
The spectrogram has a wide range of applications in speech analysis.Automatic segmentation of phonemes is a basic stage in the process of speech recognition and it will divide the speech sentences according to phoneme characteristics.This paper proposes a method of automatic segmentation of phonemes, A deformation function that represents the change of the density of the speech spectrum, and an adaptive threshold technique to locate the edge of each phoneme segment.This method is implemented on the computer.We used a set of experimental data from a spectrogram database The automatic segmentation method described in this article divides the phonemes and compares the result to a result split by a phonetician, yielding a recognition rate higher than 93%. This method, as part of a speech recognition system, is already in a speech analysis system use.