Neural Networks for Speech Application.
Abstract
This is a general introduction to the reemerging technology called neural networks, and how these networks may provide an important alternative to traditional forms of computing in speech applications. Neural networks, sometimes called Artificial Neural Systems (ANS), have shown promise for solving problems that traditional algorithmic and AI (artificial intelligence) approaches have found difficult. The world's greatest super-computer calculates Pi to thousands of decimal places in seconds using algorithmic techniques, but it may not ever be able to recognize a smiling human face when only a non-smiling version of this face is available for comparison. One reason for this is that computer process information serially, and an incredibly large number of serial steps are required to perform such a task. Therefore, even with the fastest computer, developing algorithms that can ignore unimportant differences in images and match-stored patterns with acceptable time delays is not an easy feat. The brain, on the other hand, processes information in a parallel fashion, distributing information and processing tasks throughout many neurons and their interconnections. ANS processors mimic this parallel structure and are able to outperform serial processors for certain tasks. They can also learn from their environment and are highly tolerant of internal failures.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1987
- Accession Number
- ADA191588
Entities
People
- Stephen A. Luse