Zero time windowing based severity analysis of hypernasal speech

This work analyses the hypernasal cleft palate speech for 4-point severity rating (0-normal, 1-mild, 2-moderate, 3-severe). Hypernasality results nasalization of vowels and some of voiced consonants. Addition of extra nasal peak in vicinity of first formant occurs due to nasalization. To resolve the closely spaced extra nasal peak and first formant, a high spectral resolution technique called zero time windowing (ZTW) is used. The speech signal is multiplied with a highly decaying impulse-like window of size approximately a pitch period. The loss in spectral resolution due to zero time windowing is restored by successive differentiation in frequency domain. Further the numerator of group delay is used to resolve closely spaced extra nasal peak and first formants. ZTW based spectrum captures the vocal tract changes even within an open and a closed phase of glottis and does not have the effect of pitch harmonics. Acoustic analysis of low vowel, high vowel and some voiced consonants, taken from the hypernasal speech, is done to rate the severity. The result shows no extra nasal peak in normal speech vowels. It appears in high vowel only for mild, both high and low vowels for moderate and severe, and in voiced consonants for severe only.