next up previous contents
Next: Language Modelling and Up: Language and Program Previous: Acoustic Modeling

Tools for Developing Automatic Speech Recognition Applications

Requests from Industry to develop applications for Automatic Speech Recognition (ASR) have inspired a number of projects with the objective of developing tools for rapidly customizing applications based on user requests. A baseline ASR has been built with a suitable Graphic User Interface (GUI). This system includes:

In addition to the basic ASR system, a number of other tools has been developed for:

With these tools, a system for training Air Traffic Controllers (ATC) has been built with partial contribution from ATS. The system is speaker-independent, accepts continuous speech, contains a fairly sophisticated grammar and multiple pronunciations of words, especially airline and city names that are pronounced differently by speakers of different mother tongues. A procedure based on new methods inspired by Stochastic Inference Networks (SIN) has been established for learning new word pronunciations. It is described in a paper of the 1995 IEEE International Conference on Acoustic, Speech and Signal Processing.

P. Boucher, R. DeMori, M. Galler, C. Snow



Thierry Baron
Mon Nov 13 10:43:02 EST 1995