Improved model parameter compensation methods for noise-robust speech recognition

In this paper we study model parameter compensation methods for noise-robust speech recognition based on CDHMM. First, we propose a modified PMC method where adjustment term in the model parameter adaptation is varied depending on mixture components of HMM to obtain more reliable modeling. A statedependent association factor that controls the average parameter variability of Gaussian mixtures and the variability of the respective mixtures is used to find the final optimum model parameters. Second, we propose a re-estimation solution of environmental variables with additive noise and spectral tilt based on expectation-maximization (EM) algorithm in the cepstral domain. The approach is based on the vector Taylor series (VTS) approximation. In our experiments on a speaker independent isolated Korean word recognition, the modified PMC show better performance than the Gales’ PMC and the proposed VTS is superior to both of them.