Speaker identification based on nonlinear speech models

Some of the work on speech processing has focused on modeling speech as an AM-FM signal. The success of the AM-FM model motivated us to investigate a similar nonlinear model and examine its application in speaker identification. Tests are carried out to compare the performance of the novel cyclic correlation based method with popular speaker identification methods based on cepstra. These studies show that the performance of the proposed method is comparable to the cepstrum based approach at high signal-to-noise ratio, but the former outperforms the latter under noisy conditions.