Different strategies for distribution clustering using discrete, semicontinuous and continuous HMMs in CSR

摘要: The authors present an overview of different strategies and refinements to share parameters in HMM models at distribution (state) level for continuous speech recognition, showing the advantages drawbacks kinds modeling. They compare them with sharing model level, achieving error reduction close 20%. Discrete, semicontinuous are also compared using these approaches. consider two ways smooth discrete distributions (interpolate detailed context dependent robust independent) derived from deleted interpolation co-occurrence smoothing.