Adaptation of acoustic model by using gain-adapted AR-HMM decomposition method

摘要

In a real environment, it is essential to adapt an acoustic model to variations of background noises in order to realize robust speech recognition. In this paper, we construct an extended acoustic model by combining a mismatch model with a clean acoustic model trained by using only clean speech. We assume the mismatch model conforms to a normal distribution with time-varying population parameters. The proposed method adapts the extended acoustic model to the noises by estimating the population parameters using Gaussian Mixture Model (GMM) and Gain-Adapted Auto-Regressive Hidden Markov Model (GA-ARHMM) decomposition method. We performed recognition experiments under noisy conditions using the AURORA2 database in order to confirm the effectiveness of the proposed method.