Storage Efficient Correlation Models and Clustering Algorithms for Acoustic Spectral Prototypes
Original Publication Date: 1985-Nov-01
Included in the Prior Art Database: 2005-Feb-19
Correlation models for acoustic spectral prototypes are disclosed which, by including reasonable constraints, reduce storage and computation requirements over previous models. In various speech-recognition systems, a plurality of spectral sound types are defined for each frame of speech input, a probability distribution is then formed indicating the likelihood of each sound type being the one corresponding to the frame.