Automatic Speech Recognition In Noisy Environments Using Wavelet Transform

The performance of speech recognition systems is mainly determined by the used acoustic feature extraction technique. Two techniques are known, namely the full-band approach and the multi-band approach using filter banks. Systems using either approach usually suffer from performance degradation in the presence of noise. In this paper, the multi-band approach using Wavelet transform is suggested for speaker-independent isolated word recognition in noisy environments. Moreover, it has been found that combining the acoustic features obtained using both the full-band approach and the Wavelet transformbased multi-band approach has led to an improvement in the achievable recognition rates especially under mismatched conditions at low signal-to-noise ratio situations. Key-Words: Speech Processing, Wavelet Transform and Automatic Speech Recognition.