A time-scaling approach to detect the closely spaced segments of noisy speech signals

A segmentation algorithm for noisy speech signal is introduced. The combination of the time scaling technique and binary Walsh transform is the fundamental concept of this method. This approach can also detect the gaps which are closely located among the signal components while segmenting the whole noisy waveform. The location of speech segments and gaps are determined from the reconstructed time-scaled signal. It is found that using Walsh transform at the reconstruction phase of time scaling process provides high capability in separating the signal components from narrow gaps. Although this method is less complex, the experimental results are satisfactory in the absence of a priori knowledge about speech and noise.