PSYCH 1102 Lecture Notes - Lecture 19: Speech Segmentation, Speech Perception, Phoneme
Document Summary
May be influenced by the moto programs used to create them. Speech needs to be segmented into meaningful units (words). A connectionist model might perform this task well by. Learning to predict next input, with word boundaries corresponding to poor predictions. Will almost never be at the end of an utterance. Expectation of next input (specific phoneme or boundary) Perceive the /s/ even though it"s been replaced by a cough. Da da or ba ba - was saying ba ba but the mouth movement was ga ga (integrated visual simulation with auditory - influenced how you heard the sound) Parents will translate their toddlers for people bc they have learned their kid"s speech habits / etc. Takes phonemic features and feeds into phonemes which feeds into words. Connected to feed upwards / makes sense. Bidirectional (connects features to phonemes to words) Competitive (/b/ and /k/ might be in competition with each other in the intermediate layer)