A number of issues realted to the development of speech-recognition systems with Hidden Markov Models (HMM) are discussed. A set of systematic experiments using the HTK toolkit and the TMIT database are used to elucidate matters such as the number of mixtures to use for a particular training-set size, the utility of various feature sets, the value of triphone modeling, etc.
Reference:
Barnard, E, Gouws, E, Wolvaardt, K and Kleynhans, N. 2004. Appropriate baseline values for HMM-based speech recognition. 15th Annual Symposium of the Pattern Recognition Association of South Africa, Grabouw, South Africa, 25 to 26 November 2004
Barnard, E., Gouws, E., Wolvaardt, K., & Kleynhans, N. (2004). Appropriate baseline values for HMM-based speech recognition. PRASA 2004. http://hdl.handle.net/10204/5512
Barnard, E, E Gouws, K Wolvaardt, and N Kleynhans. "Appropriate baseline values for HMM-based speech recognition." (2004): http://hdl.handle.net/10204/5512
Barnard E, Gouws E, Wolvaardt K, Kleynhans N, Appropriate baseline values for HMM-based speech recognition; PRASA 2004; 2004. http://hdl.handle.net/10204/5512 .
