论文部分内容阅读
辅音的谱结构在语音分析、合成、识别以及人工智能等方面有十分重要的意义。由于线性预测理论在语音技术中的应用日益深入，因此本文采用线性预测的方法对汉语普通话辅音进行谱分析。整个谱分析工作在通用的微机系统上进行。本文重点是解决辅音谱分析中几个特别困难的问题，如在取样方面，如何采取措施以保证原始数据的准确性；在谱分析时，对信号帧长、预测阶数如何选择。最后，通过实验，对全部汉语普通话辅音音素确定了在给定取样率情况下的信号帧长及预测阶数，作出，它们的线性预测谱。并与语图仪实测谱图比较，结果令人满意。
The consonant spectrum structure is of great significance in speech analysis, synthesis, recognition and artificial intelligence. Due to the increasing application of linear prediction theory in speech technology, this paper uses the linear prediction method to perform spectral analysis of Chinese Mandarin consonants. The entire spectrum analysis work on a common computer system. This article focuses on solving several particularly difficult problems in consonant spectrum analysis, such as how to take measures to ensure the accuracy of the original data in sampling; how to choose the signal frame length and prediction order in spectral analysis. Finally, through experiments, we determine the signal frame length and prediction order for all Mandarin Chinese consonant phonemes at a given sampling rate, and make their linear prediction spectra. The results were satisfactory compared with the spectrogram measured spectra.