THE ROLE OF EVIDENCE AND COUNTER-EVIDENCEIN SPEECH PERCEPTION

Speech perception is robust in adverse acoustic conditions when other sounds mask portions of the signal. Under experimental conditions, intelligibility can remain high even when much of the spectrum is removed, and can be further increased when noise fills the spectral gaps. These results suggest that listeners may be able to exploit noisy regions during recognition. This paper presents a joint psychophysical and modelling study into the intelligibility of band-pass filtered speech. Experiment 1 measured the intelligibility of digit sequences which were band-pass filtered at a range of centre frequencies. A model employing missing data techniques produced qualitatively similar performance to listeners. In experiment 2, listeners were presented with digit sequences filtered into two widely separated narrow bands. Band limited noise at a range of levels was added into the spectral gap between the two speech bands. A small improvement in intelligibility with increasing noise level was measured. Implications for identification metrics involving evidence and counter-evidence are discussed.

[1]  H.J.M. Steeneken,et al.  On measuring and predicting speech intelligibility , 1992 .

[2]  Phil D. Green,et al.  Auditory scene analysis and hidden Markov model recognition of speech in noise , 1995, 1995 International Conference on Acoustics, Speech, and Signal Processing.

[3]  Steven Greenberg,et al.  Speech intelligibility derived from exceedingly sparse spectral information , 1998, ICSLP.

[4]  Richard M. Warren,et al.  Auditory Perception: A New Analysis and Synthesis , 1998 .

[5]  Phil D. Green,et al.  Handling missing data in speech recognition , 1994, ICSLP.

[6]  J A Bashford,et al.  Spectral restoration of speech: Intelligibility is increased by inserting noise in spectral gaps , 1997, Perception & psychophysics.

[7]  C. J. Darwin,et al.  Chapter 11 – Auditory Grouping , 1995 .

[8]  R. M. Warren.  Perceptual Restoration of Missing Speech Sounds , 1970, Science.

[9]  Richard P. Lippmann,et al.  ROBUST SPEECH RECOGNITION WITH INTERRUPTIONS, AND NOISE:': , 1997 .

[10]  R. M. Warren,et al.  Spectral redundancy: Intelligibility of sentences heard through narrow spectral slits , 1995, Perception & psychophysics.

[11]  Phil D. Green,et al.  Some solution to the missing feature problem in data classification, with application to noise robust ASR , 1998, Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181).

[12]  Martin Cooke,et al.  Modelling auditory processing and organisation , 1993, Distinguished dissertations in computer science.

[13]  R. G. Leonard,et al.  A database for speaker-independent digit recognition , 1984, ICASSP.