基于专用矢量量化码本方案的汉语语音的四声调识别

通过对四声调在基音图形上反映出的特征所作的分析,提出了一种合理地构成四声调特征矢量的方法,使VQ在识别四声问题的应用方案得到简化。综合两种判决方法各自的特点而设计的联合判决方案,有效地消除了大部分的误识现象。实验结果表明,基于上述方案的系统对1075个不同字的四声调的识别率达到99％以上。

Chinese Four Tones Recognition Based on Special Vector Quantization Codebook Approach

This paper analyzes typical pitch graphs of words with each of the four tones of the spoken Chinese language, Certain distinguishing features are noted, A reasonable method is presented for choosing a fea- ture vector which can be used in a simplified vector quantization method in a four tones recognition system, A combination decision plan designed by the authors makes use of the advantages of two decision methods and sim- plifies distinguishing of second and third-tone words, Recognition accuracy has been increased to over 99%.