This paper describes the parallel implementation of a Hidden Markov Model (HMM) for spoken language recognition on the MasPar MP-1. A major drawback of using HMMs for speech recognition is the amount of processing time required to develop and test the model. By exploiting the massive parallelism of explicit duration HMMs, we can develop more complex models for real-time speech recognition. Implementational issues such as choice of data structures, method of communication, and utilization of parallel computation functions will be explored. The results of our experiments show that the parallelism in HMMs can be effectively exploited by the MP-1. Training that use to take more than a week can now be completed in about an hour. Once trained, the system can recognize the phones of a test utterance in a fraction of a second.

Date of this Version

June 1994