Tone Modeling for Thai Dysarthric Speech RecognitionBy Thubthong, Nuttakorn; Kayasith, Prakasith; RESNA 27th International Annual Conference 2004: Technology & Disability: Research, Design, Practice, & Policy,
Publication Date: 2004
Paper discusses a novel approach to exploit the tone features in the Thai language to improve speech recognition system performance for people with dysarthric speech. The Thai language is monosyllabic, with five lexical tones: (1) mid, (2) low, (3) falling, (4) high, and (5) rising. Studies have shown that people with dysarthric speech who may not be able to produce a variety of clear sounds of phonemes may have more control over their tone features. The researchers hope to incorporate tone information in speech recognition systems to improve speech performance for people with dysarthric speech. A quantitative analysis of the researchers approach is presented, and demonstrated that incorporating tone models improved overall recognition performance. Implications for further research are discussed.
Published by: Rehabilitation Engineering & Assistive Technology Society of North America (RESNA) (Website:http://www.resna.org)