论文部分内容阅读
本文在定量分析了声道长度与共振峰位置的关系之后,结合听觉系统中语音频谱的编码方式,提出了元音识别的模板假设,为解决非特定人语音识别问题进行了初步的探讨。根据这个假设提出了一个可行的元音模板匹配识别算法,通过非特定人语音识别实验证实了元音识别的模板假设的正确性,以及元音模板匹配识别算法的与说话人无关的特性,同时也表明该算法比利用目前主流技术中的倒谱系数的识别方法具有更好的鲁棒性。
After quantitatively analyzing the relationship between the channel length and the position of the formant, this paper puts forward the template hypothesis of vowel recognition based on the encoding method of speech spectrum in auditory system, and probes into the problem of non-specific speech recognition. Based on this assumption, a feasible vowel template matching recognition algorithm is proposed. The correctness of the vague template recognition and the non-speaker related features of the vowel template matching recognition algorithm are confirmed by the experiment of non-specific speech recognition. It also shows that this algorithm has better robustness than the recognition method using the cepstral coefficients in the current mainstream technology.