14th International Congress of Phonetic Sciences (ICPhS-14)

San Francisco, CA, USA
August 1-7, 1999

Using Neural Networks to Investigate the Relationship between Speech Production and Perception

Jörgen Pind

Department of Psychology, University of Iceland, Reykjavík, Iceland

The relationship between speech perception and speech production is a complex one. Different views have been offered on this relationship in the past. In particular, the question of the extent to which perception mirrors production has been contested. For this relationship to be fruitfully investigated it would be advantageous to adopt some objective means of deriving perceptual hypothesis from production data. This paper shows that simple neural networks can be used for this purpose. Two simulation experiments will be presented and compared with experiments dealing with the rate-dependent perception of Voice Onset Time (VOT) and of duration as a cue for vowel and consonant quantity. Results show that a neural network containing no hidden units is able to predict the result of a perception experiment dealing with vowel and consonant durations as cues to quantity, even at variable speaking rates, but fails to predict the results of an experiment dealing with the rate-dependent perception of VOT.

Full Paper

Bibliographic reference.  Pind, Jörgen (1999): "Using neural networks to investigate the relationship between speech production and perception", In ICPhS-14, 2085-2088.