Performance of an Audio Perceptron.
Abstract
Perceptrons are a class of simple adaptive pattern-recognition devices built of crude model neurons. In the work a perceptron is used to recognize patterns generated by an audio preprocessor. The preprocessor is modeled on the cochlea and cochlear ganglion of the cat with the assumption that these systems are similar to those in humans. Nonsense syllables are used as input to the preprocessor and the perceptron is taught to dichotomize the syllables through a negative-reinforcement training procedure. The preceptron is tested for its ability to learn various dichotomies as a function of the complexity of the dichotomy and as a function of the number of different voices used. It is further tested for its ability to generalize from one set of speakers to another. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1971
- Accession Number
- AD0740125
Entities
People
- Mark Gordon Scattergood
Organizations
- Cornell University