Robust Wavelet based Sub-band Features for Speech Recognition

O. Farooq (India) and S. Datta (UK)


Speech recognition, feature extraction, wavelet based processing


The most commonly used feature extraction techniques are based on short time Fourier transform. Since speech signals are not stationary even for short durations, hence STFT is not a good choice. Thus, to overcome this problem of non-stationarity in the signal wavelet based feature extraction technique has been proposed. The robustness of these features for the phoneme recognition problem is also explored and the performance is compared with the MFCC features.

Important Links:

Go Back