CONTINUOUS SPEECH DICTATION - FROM THEORY TO PRACTICE

This paper gives an overview of the Philips research system for phonem e-based, large-vocabulary, continuous-speech recognition. The system h as been successfully applied to various tasks in the German and (Ameri can) English languages, ranging from small vocabulary tasks to very la rge vocabulary tasks. Here, we concentrate on continuous-speech recogn ition for dictation in real applications, the dictation of legal repor ts and radiology reports in German. We describe this task and report o n experimental results. We also describe a commercial PC-based dictati on system which includes a PC implementation of our scientific recogni tion prototype. In order to allow for a comparison with the performanc e of other systems, a section with an evaluation on the standard Wall Street Journal task (dictation of American English newspaper text) is supplied. The recognition architecture is based on an integrated stati stical approach. We describe the characteristic features of the system as opposed to other systems: 1. the Viterbi criterion is consistently applied both in training and testing; 2. continuous mixture densities are used without tying or smoothing; 3. time-synchronous beam search in connection with a phoneme look-ahead is applied to a tree-organized lexicon.