 |
Speech Recognition and Intrinsic Variation (SRIV2006)Toulouse, France |
 |
|
Speech Recognition and Intrinsic Variation (SRIV2006)Toulouse, France |
|
It is known that the information necessary for the identification of coarticulated vowels is distributed throughout the duration of the vowels and the adjacent phonemes. This information is embedded in the speech signal in various acoustic patterns: static, dynamic, and temporal. A recent study identified seven types of acoustic patterns that might be exploited by listeners in the identification of coarticulated vowels. This paper extends the previous study and focuses on two problems. First, it presents a quantitative analysis of the underlying distribution of the acoustic information throughout vowels and adjacent consonants by employing vowel classification experiments. Second, it presents a quantitative analysis of the within-groups and between-groups sources of variability that make the total intrinsic variability of speech, and shows that the results of such analysis correlate with and predict the results obtained in the vowel classification experiments that reflect the distribution of information. The findings of this paper may be important for automatic speech recognition and suggest some basic improvements to such techniques.
Bibliographic reference. Dusan, Sorin (2006): "On the distribution of information and intrinsic variability for classification of coarticulated vowels", In SRIV-2006, 53-58.