Toward Automatic Adaptation of the Acoustic Models and of the Formulation Variants in a Directory Assistance Application

The framework of this work is an already operational voice activated Directory Assistance (DA) service that allows a large amount of data from the field to be collected. Our goal is to improve its performance by adapting the acoustic models and the formulation variants of the system to the field. For model adaptation, we propose to dynamically generate without supervision additional Hidden Markov Models tailored to the application environment and vocabulary. We report results showing significant improvements obtained in the recognition of the city names. A relevant problem in DA for business listings is that customers formulate their requests for the same listing with a great variability. We show that an unsupervised approach allows to detect user formulations that were not foreseen by the designers, and that can be added, as variants, to the denominations already included in the system to reduce its failures.