论文部分内容阅读
针对低信噪比环境 ,提出一种汉语语音声韵母切分新方法。以语音信号非线性产生机制中存在混沌特性为依据 ,将普通分形维数扩展为多尺度分形维数 ,用于考察语音信号在不同最大观测分辨率下的局部自相似性。利用稳定声韵母段及其之间过渡段在多尺度分形维数上的不同特性能较好地区分二者。由此针对汉语音节“声母 +韵母”的结构特点设计了一种简单而高效的汉语语音声韵母切分方法。在干净语音测试集下测试 ,切分正确率为 95 .2 % ;在信噪比为10 d B的噪声环境下 ,正确率达到 82 .3%。
Aiming at low signal to noise ratio environment, a new segmentation method of Chinese phonetic vowel is proposed. Based on the existence of chaos in the non-linear mechanism of speech signal, the general fractal dimension is extended to multi-scale fractal dimension to investigate the local self-similarity of speech signals under different maximum resolution. The use of stable vocal constellation and the transition between the multi-scale fractal dimension of different characteristics can better distinguish between the two. Thus a simple and efficient method of segmentation of Chinese phonetic vowel was designed according to the structural features of Chinese syllable “initials + vowels”. When tested under a clean speech test set, the correct segmentation rate was 95.2%, and the correct rate was 82.3% under a noise environment with a signal-to-noise ratio of 10 dB.