Anti-noise Algorithms Of Speech Recognition Research

In the process of speech communication, the speech signal is often corrupted acoustically by ambient noise, electronic Noise, or speech of other talkers. So a received signal is not pure, but noisy speech signal. Therefore, speech enhancement technique plays a great part of digital speech signal processing.We study the normal Subtract spectral method. With the conditions ,The page put forward to a new method of denosing based on the normal substact spectral. Noise spectral of speech silence can gained with short time average cross zero ratio , Noise spectral of other frames estimate can gained based on the Principle of Least Square Algorithm. Subtracting a noise spectral estimate from a noisy speech spectrum can therefore retrieve the spectrum of clean speech. The enhanced speech is then reconstructed via an IFFT using the modified magnitude spectrum and the original phase spectrum. The simulation result indicates that the application of objective measures, speech spectrograms, as well as subjective listening tests.Wavelet can analyze signal at time domain and frequency simultaneously, so it can be de-noise effectively, and the election of thresholds has immediate relation to the result of de-noising. So we put forward a wavelet-based denoising method by modified threshold function which can be researched adaptively satisfy the requirement of time variable signal in real-time processing, It shows the algorithm can remove the noise effectively and denoised signals can restrict the original signal very closely by experiments.Because the unstable of noise,Denoising of Subtract spectral and wavelet threshold are not ideal. So We study a new speech recognition method based on HMM and WNN by using the abilities of the modeling ability in the time domain of Hidden Markov Model and the powerful classification and decision ability of Wavelet Neural Network. The method is less to rely on noise.Compared the traditional method(HMM),It shows that this method has good performance specially in noisy speech condition by the experiment.