Adaptive control of the interaction of segmentation and pattern classification: A pilot study applied to automatic recognition of continuous speech

In the recognition of complex patterns it is becoming more apparent that segmentation (object isolation) and pattern classification cannot be performed by independent procedures. A pilot study is described which employs feedback to control the interaction of segmentation and segment classification in a system under development for automatic recognition of continuous speech. The potential advantages of employing such interaction in this work are discussed and data are presented which indicate that this approach is promising.