Under-resourced speech recognizers may benefit from data in languages other than the target language. In this paper, we report how to boost the performance of an Afrikaans automatic speech recognition system by using already available Dutch data. We successfully exploit available multilingual

. While the method has been developed for and suc-cessfully implemented in the IBM Real Time Speech Recognizers, its generality makes it applicable in other areas where the problem of es-timating probabilities from sparse data arises. Sparseness of data is an inherent property of any real text

community. This paper describes an effort to implement a speaker dependent continuous speech recognizer for a less resourced non-Latin language, namely, Sinhala. Using readily available open source tools, it shows that fairly accurate speaker dependent ASR systems for continuous speech can be built

JANUS [17] was designed for the translation of spontaneous human-to-human speech. Before the 1994 CSR evaluation, JANUS was run with vocabularies of up to 2500 words. JANUS was also tested on the Conference Registration and the Resource Management tasks. The best error rate on the '89 Resource

“Deep ” architecture in human speech recognition

intuitively meaningful figures of merit: 1) the noise level necessary to achieve comparable human performance and 2) the deviation of the pattern of confusions from human performance. Examples are given of recognizers evaluated in this way, and the role of these performance measures in automatic speech

Creating a speech recognizer from scratch can be quite daunting for the beginner. We present a simple and economical method for creating speech recognizers. Our goal is a favorable cost/benefit ratio rather than the best possible recognition score. We are mainly interested in recognizers

In this contribution we examine large speech corpora of pre-pared broadcast and spontaneous telephone speech in Ameri-can English and in French. Starting with the question whether ASR systems behave differently on male and female speech, we then try to find evidence on acoustic-phonetic, lexical

This thesis is about modeling, analyzing, and predicting errorful behavior in large vocabulary continuous speech recognition systems. Because today's state-of-the-art recognizers are not designed to be situated naturally in an error feedback loop, they are ill-positioned for inclusion in multi

Current speech recognition systems require large amounts of transcribed data for parameter estimation. The transcription, however, is tedious and expensive. In this work we describe our experiments which are aimed at training a speech recognizer with only a minimal amount (30 minutes