Join the P-interest workshop today at noon in the Greenberg Room, where Santiago Barreda (UC Davis) will speak about speaker-adaptive vowel perception.
Modelling speaker-adaptive vowel perception using a statistical pattern-recognition model
In this talk I will outline experimental evidence suggesting that vowel perception and the determination of apparent speaker characteristics are related processes, and that they interact and cooperate. In this view of vowel perception, the interpretation of acoustic information is based on what the listener expects for a given speaker, and the detection of speaker changes is an important aspect of speech perception. This approach to speech perception is able to explain a wide range of experimental results including the influence of instructions on vowel perception, the increased reaction times associated with mixed-speaker listening situations, and the indirect effect of some speech cues (e.g. pitch) on perceived vowel quality. I will outline a statistical model of vowel perception that identifies vowel sounds probabilistically, on the basis of speaker-specific expectations. The results of some behavioral experiments were simulated using this model to compare the predictions made by alternative views of vowel perception. Results support the notion that vowel perception is tied to speaker expectations and the detection of speaker changes rather than being deterministically related to the acoustic characteristics of a speech sound.