BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Principles of AI-driven Neuroscience and Translational Biomedicin
 e - Mate Aller
DTSTART:20260310T160000Z
DTEND:20260310T170000Z
UID:TALK243511@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:Title: Efficiency and flexibility in an LSTM model of human sp
 oken word recognition\n\n \n\nAbstract: Recent advances in artificial neur
 al networks have enabled the design of automatic speech recognition system
 s that identify spoken words with an accuracy approaching human listeners.
   By analysing the functional characteristics and internal representations
  of such systems\, and comparing them to human listeners\, we can gain nov
 el insights into classic psycholinguistic findings and testable prediction
 s for neuroimaging experiments exploring the neural computations for human
  speech perception.\n\nHere we build on a recently published end-to-end mo
 del of human speech recognition (‘EARSHOT’). This is a recurrent (LSTM
 ) neural network trained to map from acoustic representations of spoken wo
 rds to representations of word meaning (semantics). It exhibits a human-li
 ke time course of word identification with parallel activation (due to pho
 nological overlap) of onset-aligned ‘cohort’ neighbours (e.g. chain/ch
 ange)\, and reduced\, or delayed competition effects between rhyme neighbo
 urs (e.g. chain/gain).\n\nWe systematically characterised EARSHOT’s beha
 viour in recognising speech across different talkers\, speaking rates\, an
 d levels of spectral detail. In addition\, we analysed the model’s hidde
 n-state dynamics to provide a mechanistic explanation for several of the o
 bserved behavioural patterns.\n\n\n \n
LOCATION:LT1
END:VEVENT
END:VCALENDAR