文档介绍：IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL. 6, NO. 1, JANUARY 1998 71 Speaker Clustering and Transformation for Speaker Adaptation in Speech Recognition Systems Mukund Padmanabhan, Member, IEEE, Lalit R. Bahl, Fellow, IEEE , David Nahamoo, Member, IEEE, and Michael A. Picheny, Member, IEEE Abstract— A speaker adaptation strategy is described that is based on ?nding a subset of speakers, from the training set, who are acoustically close to the test speaker, and using only the data from these speakers (rather than plete training corpus) to reestimate the system parameters. Further, a linear transformation puted for every one of the selected training speakers to better map the training speaker’s data to the test speaker’s acoustic space. Finally, the system parameters (Gaussian means) are reestimated speci?cally for the test speaker using the transformed data from the selected training speakers. Experiments showed that this scheme is capable of providing an 18% relative improvement in the error rate on a large-vocabulary task with the use of as little as three sentences of adaptation data. Index Terms— Data transformation, speaker adaptation, speaker clustering. I. I NTRODUCTION I N THE LAST few years, several advances have been made in improving the error rate of continuous-speech- recognition systems [1]. For instance, the best word-error rates on test data drawn from the Wall Street Journal ( WSJ) data base—as reported by different participants in the WSJ task [1]—hover in the neighborhood of 7–8% for large-vocabulary speaker-independent systems. Though this represents a reason- able level of performance on this particular test data, there is still scope for further improvement. One way to improve the performance of these systems is to make the system parameters speaker dependent. However, large-vocabulary systems tend to have a large number of parameters, and in order to robustly estimate these parameters, a large amount of training data is needed. This