An integrated approach to feature compensation combining particle filters and hidden Markov models for robust speech recognition

摘要

Obtaining accurate hidden Markov model (HMM) state sequences is a research challenge to warrant good system performance in particle filter (PF) compensation for noisy speech recognition. Instead of using specific knowledge at the model and state levels which is hard to estimate, we pool model states into clusters as side information. Since each cluster encompasses more statistics when compared to the original HMM states, there is a higher possibility that the newly formed probability density function at the cluster level can cover the underlying speech variation to generate appropriate PF samples for feature compensation. Testing the proposed PF-based compensation scheme on the Aurora 2 connected digit recognition task, we achieve an error reduction of 12.15% from the best multi-condition trained models using this integrated PF-HMM framework to estimate the cluster-based HMM state sequence information.