FACTORIZED HIDDEN LAYER ADAPTATION FOR DEEP NEURAL NETWORK BASED ACOUSTIC MODELING

Abstract

The automatic speech recognition (ASR) technology have become widely popular since the deep neural networks (DNNs) have been integrated into ASR systems. However, DNNs are susceptible to performance degradations due to the mismatches between the training and testing conditions. Therefore, adaptation techniques are used to reduce this mismatch. This thesis proposes two methods to adapt the DNN acoustic models for ASR. Namely, factorized hidden layer (FHL) and subspace learning hidden unit contributions (LHUC). FHL aims at modeling speaker dependent (SD) hidden layers by representing an SD affine transformation as a linear combination of bases. The combination weights can be initialized using speaker representations like i-vectors and then reliably refined in an unsupervised adaptation fashion. The subspace LHUC improves the LHUC method by constructing a subspace to estimate the SD parameters. Experiments have shown that the both methods improve the ASR performance significantly over baselines and other state-of-the-art DNN adaptation approaches.