Application of Kohonen self-organising feature maps to smoothing parameters of hidden Markov models for speech recognition

The paper introduces a new method to smooth the parameters of hidden Markov models (HMMs) which produces improved recognition results when only a limited amount of training data is available. The method uses the Kohonen self-organising feature map (KSOFM) as a clustering technique in codebook design for discrete HMMs. Comparison of the new smoothing method with the smoothing method based on k-means clustering is made