Statistical methods for speech recognition
Statistical methods for speech recognition
Hi-index | 0.00 |
A formal structural representation of speech is presented in this paper. The representation is developed within the Evolving Transformation System (ETS) formalism and encapsulates speech processes at the articulatory level. We show how the class structure of several consonantal phonemes of English can be expressed via articulatory gestures. Elements of these classes can be detected in a corresponding structural representation of continuous speech. Our experimental results on the MOCHA articulatory corpus support the hypothesis that the proposed articulatory representation captures sufficient information for the accurate structural identification of phonemic classes.