Current speech recognition algorithms designed for use in severe noise environments have demonstrated robust performance in ambient noise levels up to 97 dB for vocabularies of approximately 30 words. Robustness with respect to human speech variability due to physiological and psychological stress is still considered a limitation to widespread application of speech recognition for applications that can induce such stress. Testing for robust speech recognition performance must assess other noise environments and must address rejection accuracy as well as recognition accuracy. A rigorous measure of rejection accuracy is obtained by submitting an algorithm to tokens that differ from legal vocabulary words by a single phoneme (speech sound). The two studies reported here addressed these issues for speaker-dependent, connected speech, template matching algorithms.
Testing for robust speech recognition performance
Performance-Tests fuer Algorithmen der robusten Spracherkennung
1986
9 Seiten, 4 Bilder, 1 Tabelle, 12 Quellen
Aufsatz (Konferenz)
Englisch
Robust Speech Recognition System for Avionics
Online Contents | 2004
|Extended VTS for noise-robust speech recognition
TIBKAT | 2009
|Noise Robust Speech Recognition Using Prosodic Information
Springer Verlag | 2005
|Is Speech Recognition Technology Robust for Older Populations?
British Library Conference Proceedings | 1999
|Noise robust speech recognition and coding in car environment
Kraftfahrwesen | 1991
|