Comparison of model and neural responses to speech

• Published in: The Journal of the Acoustical Society of America Vol. 77; no. S1; p. S80
• Main Author: Payton, Karen L.
• Format: Journal Article
• Language: English
• Published: 01-04-1985
• ISSN: 0001-4966; 1520-8524
• DOI: 10.1121/1.2022529

This paper describes a computer simulation of vowel processing by the auditory periphery. Models of functionally separate parts of the periphery were modified and concatenated to form a composite model. The model accepts digitized temporal waveforms as inputs and calculates a population of neural rate functions. Subsequently, the model's average and synchronized rate responses are calculated. The model encompasses efforts by Guinan and Peake [MIT R-LE, QPR 84, 320–326 (1967)], Allen and Sondhi [J. Acoust. Soc. Am. 66, 123–132 (1979)], Allen [J. Acoust. Soc. Am. 68, 1660–1670 (1980)], and Smith and Brachman [Biol. Cybernet. 44, 107–120 (1982)]. Synthesized vowels were used as stimuli after an evaluation with pure tones. Results indicate that calculated neural responses, unlike simulated basilar membrane responses, are strongly synchronized only to formant peaks. The formant peaks are preserved in the population average rate data only at low stimulus levels and at onset, while a synchronized reponse represents formant peaks well at all stimulus levels tested. Results will be compared to neural population data [M. B. Sachs and E. D. Young, J. Acoust. Soc. Am. 68, 858–875 (1980)]. [Work supported by NSF.]