Models for Speech Signals Analysis Taking into Account the Effect of Masking

An auditory model for speech data reduction is outlined. The Fourier spectrum is calculated for successive millisecond segments and the power in successive critical bands (Zwicker and Feldtkeller) is determined. In the next stage conditions of masking within a segment is determined and also the forward masking within 100 to 200 ms and the backward masking within 25 ms. Frequency-time quanta which remain unmasked constitute the final specification. The average number of significant quanta within a sample is 3 to 10.