论文部分内容阅读
针对非限定文本的自动讲话人辨认问题,对提取有效的识别特征进行了理论和实验分析,使用语音信号LPC倒谱的瞬时谱和过渡谱的特征参量作为识别特征,运用矢量量化技术,设计并建立了19人组成的ASI系统,为每人分别建立了“动态”和“静态”倒谱系数的两种码本,在连续语音完全不限文本的情况下进行系统的识别实验,得到了较理想的结果。实验表明,上述两种特征矢量均为非常有效的识别特征且可互补结合运用于识别,使性能进一步提高。本文同时还对简单信道变化情况予以初步考察,得到一些有意义的结果。
In order to solve the problem of automatic speech recognition for unrestricted texts, theoretical and experimental analysis of effective recognition features is made. Using the instantaneous spectrum of the speech signal LPC cepstrum and the feature parameters of the transition spectrum as the recognition features, vector quantization techniques are used to design An ASI system composed of 19 persons was established. Two kinds of codebooks of “dynamic” and “static” cepstral coefficients were set up for each person, and the system identification experiment was carried out under the condition of continuous speech without any limitation. The ideal result. Experiments show that the above two kinds of feature vectors are very effective identification features and can be combined complementarily for recognition, so as to further improve the performance. In the meantime, we also make a preliminary investigation on the changes of simple channels and get some meaningful results.