Tree-based state clustering for large vocabulary speech recognition

The key problem to be faced when building a HMM-based continuous speech recogniser is maintaining the balance between model complexity and available training data. For large vocabulary systems requiring cross-word context dependent modelling, this is particularly acute since many such contexts will never occur in the training data. This paper describes a method of creating a tied-state continuous speech recognition system using a phonetic decision tree. Results are presented for the Resource Management and Wall Street Journal tasks where very good performance is achieved. The method is compared to a traditional model-based procedure and shown to be clearly superior