Vowel-like region based acoustic- phonetic analysis for phone recognition
No Thumbnail Available
In landmark-based speech recognition approach, acoustic-phonetic information is extracted from the regions around certain landmarks. In a similar direction, a vowel-like region (VLR) detection based approach is proposed in this thesis for phone recognition, where VLRs are considered as landmarks. Vowels, semivowels and diphthongs are treated as VLRs and the rest category of sounds as non-vowel-like regions (non-VLRs). Basic signal characteristics of VLR and non-VLR are di erent. VLRs are used as carriers with one or more non-VLRs supporting around them. The VLRs are produced by exciting the vocal tract that has a wide open con guration or moderate constriction. On the other hand, in the production of non-VLRs, the vocal tract has a narrow constriction or a complete closure con guration. Hence, it is not proper to treat VLRs and non-VLRs in a similar fashion by computing same set of features out of them.
Supervisor: S. R. Mahadeva Prasanna
ELECTRONICS AND ELECTRICAL ENGINEERING