Self-learning speaker adaptation based on spectral bias source decomposition, using very short calibration speech

摘要: A speaker adaptation technique based on the separation of speech spectra variation sources is developed for improving speaker-independent continuous recognition. The include acoustic characteristics, and contextual dependency allophones. Statistical methods are formulated to normalize characteristics then adapt mixture Gaussian density phone models phonologic characteristics. Adaptation experiments using short calibration (5 sec./speaker) have shown substantial performance improvement over baseline recognition system.