Segmentation and labeling of speech: a comparative performance evaluation.
Henry G. Goldberg · 1976
release and sale; its distribution is unlimited. This thesis is a study of speech recognition at the parametric level. [t attempts to evaluate and understand the relative merits of a number of alternative design choices at that level. Such a study raises issues in Artificial]ntelligence, Linguistics, Acoustics, Pattern Recognition, Statistics, and Speech Understanding research. In particular _ it involves an investigation of segmentation and labeling techniques, and the use of parametric representations for the acoustic signal in those techniques. Every speech recognition system employs some parametric representation and some initial signal to " symbol transformation. We show the performance currently available for these initial processes, and assert that such performance is comparable to human performance. We present the relative merits of some typical parametric representations, and develop a methodology for such comparative evaluation. Simple, parameter-independent schemes for segmenting, labeling, and training are developed as well. The role of pattern classification techniques is clarified, as it relates to the initial signal to symbol transformation..