Real-Time Bayesian Inference: A SoftComputing Approach to EnvironmentalLearning for On-Line Robust AutomaticSpeech Recognition

摘要

In this paper, we developed soft computing models for on-line automaticspeech recognition (ASR) based on Bayesian on-line inference techniques. Bayesianon-line inference for change point detection (BOCPD) is tested for on-line environ-mental learning using highly non-stationary noisy speech samples from the Aurora2speech database. Significant improvement in predicting and adapting to new acous-tic conditions is obtained for highly non-stationary noises. The simulation resultsshow that the Bayesian on-line inference-based soft computing approach would beone of the possible solutions to on-line ASR for real-time applications.