Research of STRAIGHT Spectrogram and Difference Subspace Algorithm for Speech Recognition

Spectrogram is of advantage to express speech signal than other features in time-domain or frequency-domain alone, combining the merit of both domains and showing the relationship of time, frequency, and amplitude directly. STRAIGHT spectrogram is a kind of spectrogram with super performance which clears up cyclical nature related to fundamental frequency from short-time spectrum. In this paper we discuss the theory of STRAIGHT and its application to Chinese digit recognition, making full use of the advantages of STRAIGHT spectrogram and difference subspace algorithm. Experiments show that the algorithm proposed is effective, yielding high recognition rate. We also list the problem of this method and future work to be done.