Wavelet Transform Using Neyman-Pearson Criterion in Speech Recognition

To obtain a high robust of speech recognition for noisy conditions, a new pre-processing stage based on wavelet thresholding algorithm is proposed in this paper. The purpose of using the DWT is to benefit from its localization property in the time and frequency domains. Compromise function is proposed compared with hard and soft thresholding function. A new thresholding value, Neyman-Pearson criterion is proposed compared with the commonly used Sqtwolog, Rigrsure, minimaxi criterion. MSE and SNR are given to evaluate the improvement of noisy speech recognition performance. The result shows that the Neyman-Pearson criterion can get a better performance especially at adverse conditions.