Using phase characteristics of speech to distinguish /m/ and /n/
Donald L. Bitzer, Robert D. Rodman, David F. McAllister, Jamie Taylor · 2006
A method is presented for extracting a useful phase signal from natural input speech. The phase signal is then processed to extract signatures for /m/ and /n/ based on the variance of the phase. An average signature is created for each of 8 speakers using three utterances of each /m/ and /n/. This average signature is then used to classify unknown input utterances. Classification accuracy averages 75%.