Modeling of output probability distribution to improve small vocabulary speech recognition in adverse environments

This paper presents a solution to the adverse environment, open microphone problem, by using the information stored in HMM output probability distributions to obtain a confidence measure of the results. This information can also be used to perform a secondary classification and improve recognition results. The system was tested on data from the TI46 database that had been corrupted by noise from the NOISEX-92 database, as well as on real-world data, and shows promising results.

[1]  Arnon D. Cohen,et al.  Robust isolated word recognition using WSP-PMC combination , 1997, EUROSPEECH.

[2]  W. W. Peterson,et al.  The theory of signal detectability , 1954, Trans. IRE Prof. Group Inf. Theory.

[3]  Chin-Hui Lee,et al.  Application of hidden Markov models for recognition of a limited set of words in unconstrained speech , 1989, International Conference on Acoustics, Speech, and Signal Processing,.

[4]  Richard Lippmann,et al.  Improving wordspotting performance with artificially generated data , 1996, 1996 IEEE International Conference on Acoustics, Speech, and Signal Processing Conference Proceedings.

[5]  Douglas A. Reynolds,et al.  Robust text-independent speaker identification using Gaussian mixture speaker models , 1995, IEEE Trans. Speech Audio Process..

[6]  Francisco Javier Caminero Gil,et al.  On-line garbage modeling with discriminant analysis for utterance verification , 1996, Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96.

[7]  Yves Normandin,et al.  Detection of ambiguous portions of signal corresponding to OOV words or misrecognized portions of input , 1996, Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96.