Acoustic Discrimination between Difficult Sound Pairs in a Single Speaker,
Abstract
Various discrimination algorithms were applied to samples of spectral energies in five 1-kHz bands of 25.6 ms frames of the difficult sound pair (f) - theta produced by a single speaker, with scores ranging between 60 and 89 percent, depending on whether single or multiple frames were used to identify segments and on whether frames were drawn from full-length segments or from middle thirds of segments. A multivariate statistical description permits probability estimates to be attached to individual recognition decisions. Aside from its use in assisting discrimination, the statistical description also characterizes the sounds in terms of their mean behavior and their sources of variability.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1976
- Accession Number
- ADA027918
Entities
People
- David J. Broad