Adaptation of the HMM distributions: Application to a VQ codebook and to a noisy environment

The authors report on the application of the adaptation of parameters of HMM distributions to a vector quantization (VQ) codebook and a noisy environment after a short training session. The adaptation to the codebook was applied to an isolated word recognition task with the 26 words of the English phonetic alphabet in speaker-independent experiments. The noise adaptation technique was applied to an isolated digit recognition task in a car. In both applications the adaptation technique significantly improved the recognition performance