Speaker Adaptation Techniques for Speech Recognition with a Speaker-Independent Phonetic Recognizer

摘要

We present a new method that improves the performance of the speech recognition system with the speaker-independent (SI) phonetic recognizer. The performance of the speech recognition system with the SI phonetic recognizer is worse than that of the speaker dependent system due to the mismatch between the training utterances and a set of SI models. A new training method that iteratively estimates the phonetic templates and transformation vectors is presented to reduce the mismatch using speaker adaptation techniques. The stochastic matching methods are used to estimate the transformation vectors for speaker adaptation. The experiment performed over actual telephone line shows that a reduction of about 40% in the error rates could be achieved as compared to the conventional method.