Improved Bayesian Training for Context-Dependent Modeling in Continuous Persian Speech Recognition

Context-dependent modeling is a widely used technique for better phone modeling in continuous speech recognition. While different types of context-dependent models have been used, triphones have been known as the most effective ones. In this paper, a Maximum a Posteriori (MAP) estimation approach has been used to estimate the parameters of the untied triphone model set used in data-driven clustering. The use of better prior parameters derived from two sets of more reliably trained biphone models has helped in this process. The result is better parameter tying where the tied-state triphone system built in this manner outperforms a similar system in which ordinary Maximum Likelihood (ML) approach was used to estimate the untied triphone system parameters. The technique may also be useful in other tying schemes used in context-dependent modeling.