Abstract | ||
---|---|---|
The Speech Training, Assessment, and Remediation (STAR) system is intended to assist Speech and Language Pathologists in treating children with articulation problems. The system is embedded in an interactive video game that is set in a spaceship and involves teaching aliens to "understand" selected words by spoken example. The sequence of events leads children through a series of successively more difficult speech production tasks, beginning with CV syllables and progressing to words/phrases. Word selection is further tailored to emphasize the contrastive nature of phonemes by the use of minimal pairs (e.g., run/won) in production sets. To assess children's speech, a discrete hidden Markov model recognition engine is used(1). Phone models were trained on the CMU Kids database(2). Performance of the HMM recognizer was compared to perceptual ratings of speech recorded from children who substitute /w/ for /r/. The difference in log likelihood between /r/ and /w/ models correlates well with perceptual ratings of utterances containing substitution errors, but very poorly for correctly articulated examples. The poor correlation between perceptual and machine ratings for correctly articulated utterances may be due to very restricted variance in the perceptual data for those utterances. |
Year | Venue | Keywords |
---|---|---|
2000 | INTERSPEECH | speech production,hidden markov model |
Field | DocType | Citations |
Interactive video,Speech training,Computer science,Speech recognition,Correlation,Phone,Artificial intelligence,Natural language processing,Hidden Markov model,Speech production,Perception | Conference | 24 |
PageRank | References | Authors |
2.89 | 6 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
H. Timothy Bunnell | 1 | 112 | 19.91 |
Debra M. Yarrington | 2 | 24 | 2.89 |
James Polikoff | 3 | 75 | 10.17 |