Peaks extraction from auditory scale spectrum based on wavelet transform

Our auditory system is more sensitive to peaks in the speech spectrum, and less sensitive to valleys in the spectrum. Many methods have been used to simulate this kind of physiological function, such as cepestral lifting in the log spectral domain, lateral inhibition, etc. In this paper, a new method is proposed to simulate this physiological function based on the wavelet transform and an auditory model. From this method, peaks can be easily extracted and enhanced. Features extracted by this model can be used as robust candidates for speech recognition.