Joint-space adaptation technique for robust continuous speech recognition

The automatic speech recognition systems experience serious performance degradation when they are deployed in the environments which are mismatched to the training environment. This dissertation introduces a maximum likelihood joint-space adaptation technique for continuous speech recognition under mismatch environments. The proposed method reduces the mismatch based on iteratively combining the N-Best hidden Markov model (HMM) inversion for feature-space adaptation and the functional transformation for model-space adaptation. In this joint-space adaptation process, the N-Best HMM inversion frame-by-frame adapts the speech features non-parametrically to compensate the temporal deviation, while the models are transformed parametrically to catch the global characteristics of the mismatch. The proposed joint-space adaptation provides a better compensation to the mismatch than either of the single-space adaptations does. In addition, this proposed method operates only on the given testing speech and the models, therefore no stereo or adaptation data are required. The performance of the proposed technique is investigated in the presence of different mismatch sources.