Neural Networks in Automatic Speech Recognition
Françoise Beaufays, Hervé A. Bourlard, Horacio Franco, Nelson H. Morgan · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · 2000
Automatic speech recognition (ASR), the technology that allows computer systems to transcribe speech waveforms into words, relies essentially on traditional digital signal processing and statistical modeling methods to analyze and model the speech signal. The core ASR technology is typically not based on connectionnist methods, even though neural network processing is commonly seen as a promissing alternative to some of the current algorithms. Because of the maturity of the current technology, neural networks have to compete with high performance algorithms to gain acceptance in the speech recognition eld, and it is only recently that signicant performance improvements over state-of-the-art systems have been obtained by using neural networks in specic subsystems of the speech r