Publications

In this paper we examine various ways to derive confidence measures for a language identification system, using phone recognition followed by language models, and describe the application of an evaluation metric for measuring the "goodness" of the different confidence measures. Experiments are conducted on the 1996 NIST Language Identification Evaluation corpus (derived from the Callfriend corpus of conversational telephone speech). The system is trained on the NIST 96 development data and evaluated on the NIST 96 evaluation data. Results indicate that we are able to predict the performance of a system and quantitatively evaluate how well the prediction holds on new data.

This paper studies the structure of foreign-accented read English speech. A system for accent identification is constructed by combining linguistic theory with statistical analysis. Results demonstrate that the linguistic theory is reflected in real speech data and its application improves accent identification. The work discussed here combines and applies previous research in language identification based on phonemic features [1] with the analysis of the structure and function of the English language [2]. Working with phonemically hand-labelled data in three accented speaker groups of Australian English (Vietnamese, Lebanese, and native speakers), we show that accents of foreign speakers can be predicted and manifest themselves differently as a function of their position within the syllable. When applying this knowledge, English vs. Vietnamese accent identification improves from 86% to 93% (English vs. Lebanese improves from 78% to 84%). The described algorithm is also applied to automatically aligned phonemes.