Paper Info
Title
Hypernasality Detection Using Zero Time Windowing
Abstract
The hypernasality in cleft palate speech is characterized by the presence of nasal peak in the vicinity of first formant of vowel spectrum. A high spectral resolution technique, which can resolve these two peaks, is desirable for the automatic detection of hypernasality. This work uses the zero time windowing (ZTW) technique for the hypernasality detection. In this technique, the speech signal is windowed with a highly decaying impulse-like window of approximately a pitch period size. The technique gives the instantaneous vocal tract spectrum free from the pitch and harmonics effect. The spectral resolution loss due to short windowing is restored by the successive differentiation in frequency domain. The numerator of group delay is used to resolve closely spaced nasal peak and first formant. The cepstral feature is extracted from the instantaneous spectrum and is used for the automatic detection of hypernasality using SVM classifier. The accuracy of classification are 76.51% for the vowel /a/ and 80.36% for the vowel /i/. The accuracy further increases when the proposed feature is fused at score level with the Mel-frequency cepstral coefficient (MFCC) feature.
Year
DOI
Venue
2018
10.1109/SPCOM.2018.8724392
2018 International Conference on Signal Processing and Communications (SPCOM)
Keywords
Field
DocType
Feature extraction,Microsoft Windows,Frequency-domain analysis,Signal resolution,Delays,Task analysis,Cepstral analysis
Computer vision,Computer science,Artificial intelligence
Conference
ISSN
ISBN
Citations 
2474-9168
978-1-5386-3821-7
0
PageRank 
References 
Authors
0.34
0
3
Name
Order
Citations
PageRank
Akhilesh Kumar Dubey101.01
S. R. Prasanna2889.30
S. Dandapat326128.51