摘要
为模拟基底膜对语音信号的分析，本文提出了一种类似小波变换的时频分析方法。该方法克服了ＳＦＴ（短时Ｆｏｕｒｉｅｒ变换）分析对高、低频段具有相同的时间分辨率和频率分辨率的缺点，弥补了小波变换只能粗糙地模拟基底膜带通滤波器特性的不足。在识别阶段，本文采用了多模糊状态综合处理的识别模式，该模式有利于对各频段统计特征参数的提取和加强。实验表明，该模型具有良好的鲁棒性，较好地模拟了人的听觉系统对语音的识别过程。
A new method of the time frequency analysis based on wavelet transform is proposed to mimic the bandpass filter′s feature of basilar membrane. The time frequency analysis remedies the defect that short Fourier translation (SFT) has the same time resolution and frequency resolution to different frequency ranges. The multifuzzy state is used as a kind of recognition pattern to the parameters. As the recognition pattern of multifuzzy state, it′s convenient to extract and enhance new characteristic parameters. Some experimental results show that representation of speech in the auditory system is more effective and has better robustness.
出处
《数据采集与处理》
CSCD
1999年第2期157-162,共6页
Journal of Data Acquisition and Processing
