On-line garbage modeling with discriminant analysis for utterance verification

Out-of-vocabulary (OOV) utterance detection and rejection are specially important and difficult problems in large-vocabulary and continuous speech recognition. C de la Torre et al. (1996) proposed an utterance verification procedure based on the use of frame-by-frame best acoustic state scores instead of using explicit garbage models. This procedure is usually referred to as online garbage modeling. In this paper, we extend our previous work in two major directions: (a) we analyze, through the use of discriminant analysis, the possibilities of using L-best local scores and N-best utterance hypothesis scores for utterance verification; and (b) we present experimental results not only for a spontaneously-spoken natural number recognition task, but also for a flexible large-vocabulary recognition task. All the results, based on a telephone database, show that the proposed online garbage modeling procedure outperforms (both in terms of performance and computational cost) other approaches based on the use of explicit garbage models.