[Colloquium] 3/11 AT 9:30AM! - Talks at TTIC: Liang Lu, University of Edinburgh

[Mary Marre]

PLEASE NOTE: TIME & ROOM CHANGE !  (for 3/11 only)

When:     Friday, March 11th at *9:30 am *

Where:    TTIC, 6045 S Kenwood Avenue, 5th Floor, *Room 530*

Who:            Liang Lu, University of Edinburgh

Title:            Deep Learning for End-to-End Speech Recognition


Abstract:

Deep learning has significantly advanced speech recognition in the past few
years. However, state-of-the-art speech recognition systems still rely on
the hidden Markov model (HMM) for sequence modelling, which has been used
for decades for speech recognition. While speech recognition is essentially
a sequence-level classification problem, HMMs convert this problem into a
frame-by-frame classification task using the hidden states under the
conditional independence assumption - a well known weakness of HMMs.
Besides, the HMM-based speech recognition pipeline is composed by a few
loosely connected modules (i.e., acoustic model, lexicon model and language
model), which are not trivial to be trained jointly.

In this talk, I will discuss the recent effort in the speech community
toward end-to-end speech recognition, and present our own work in this
area. In particular, I will discuss three models that directly compute the
conditional probability of the target sequence (words, phonemes or letters)
given the source sequence (acoustic frames) without using HMMs. Firstly,
the Connectionist Temporal Classification (CTC) proposed by Alex Graves a
few years ago, which has recently been adopted and further developed by
Google. Secondly, attention-based recurrent neural network encoder-decoder
originally proposed for machine translation by researchers from Yoshua
Bengio's group. And finally, segmental recurrent neural network - a
combination of segmental conditional random field with encoder recurrent
neural network.


Host: Karen Livescu, klivescu at ttic.edu



Mary C. Marre
Administrative Assistant
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 504*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20160310/934e816f/attachment.htm