Automatic speech recognition based on adaptation and clustering using temporal-difference learning

This paper describes a novel approach based on online unsupervised adaptation and clustering using temporal-difference (TD) learning. Temporal-difference learning is a reinforcement learning technique and is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. The adaptation progresses based on rewards that represent correctness of outputs. The adapted models gradually accumulate and cluster with the environmental conditions and can immediately adapt by selecting the optimal model from the clusters. We conducted speech recognition experiments by a connected digit recognition in noisy environments including the variation of speakers and noises. The results verify that the proposed method has a higher recognition performance than the conventional adaptation method.