ASA 125th Meeting Ottawa 1993 May

3pSP8. A comparison of synchronous and asynchronous analysis of speech.

Grant J. Jensen

William J. Strong

Dept. of Phys. and Astron., Brigham Young Univ., Provo, UT 84602

Speech having a high signal-to-noise ratio and digitized at 16 kHz was used for the comparison. An ``ideal'' synchronous analysis was obtained by handmarking the speech events---voicing periods, plosive bursts, and so on. An asynchronous analysis was obtained from 25-ms Hamming windowed segments of speech stepped at 7.5-ms intervals. In both cases the speech sample---event or windowed segment---was Fourier transformed and displayed in a ``trispectral'' space. The synchronous analysis produced erratic event-to-event spectra for short voicing periods. The asynchronous analysis produced erratic segment-to-segment spectra for long voicing periods. It also tended to smear short-duration, transitional features. Various examples of trispectral plots obtained from the two methods will be shown for comparison.