The ''neural'' phonetic typewriter
Recently, researchers have placed great hopes on artificial neural networks to perform such ''natural'' tasks as speech recognition. This was indeed one motivation for us to start research in this area many years ago at Helsinki University of Technology. This article describes the result of that research - a complete ''neural'' speech recognition system, which recognizes phonetic units, called phonemes, from a continuous speech signal. Although motivated by neural network principles, the choices in design must be regarded as a compromise of many technical aspects of those principles. As our system is a genuine ''phonetic typewriter'' intended to transcribe orthographically edited text from an unlimited vocabulary, it cannot be directed compared with any more conventional, word-based system that applies classical concepts such as dynamic time warping and hidden Markov models.
- Research Organization:
- Helsinki Univ. of Technology (FI)
- OSTI ID:
- 5263327
- Journal Information:
- Computer; (United States), Journal Name: Computer; (United States) Vol. 21:3; ISSN CPTRB
- Country of Publication:
- United States
- Language:
- English
Similar Records
Neural speech recognition: continuous phoneme decoding using spatiotemporal representations of human cortical activity
Application of backpropagation neural networks to phonetic element classification