[Colloquium] Talks at TTIC: Bahador Nooraei, TTIC Student

[Dawn Ellis]

When:     Tuesday, September 2nd at 3pm

Where:    TTIC, 6045 S Kenwood Avenue, 5th Floor, Room #526

Who:       Bahador Nooraei, TTIC Student

Title:       Deep Canonical Correlation Analysis for Multi-View Feature
Learning in
              Speech Recognition

Abstract:

Canonical Correlation Analysis (CCA) has been used to learn acoustic
features when a second view of data is available during training. When the
second view is articulatory measurements, the new learned features have
been shown to improve performance in various speech recognition tasks,
including phonetic recognition and frame classification. A new method, Deep
CCA, was recently proposed as a non-linear extension of CCA. In Deep CCA,
each view of data is transformed using a Deep Neural Network. The two
networks are jointly trained to maximize the canonical correlation of their
output vectors.

In this talk we review CCA and Deep CCA and their application to multi-view
feature learning for speech recognition. We will also present the latest
results from our ongoing experiments on the Wisconsin X-ray Microbeam
Database. The results are encouraging and show that features learned using
the Deep CCA method improve phonetic recognition performance.

Adviser:  Karen Livescu,  klivescu at ttic.edu

-- 
*Dawn Ellis*
Administrative Coordinator,
Bookkeeper
773-834-1757
dellis at ttic.edu

TTIC
6045 S. Kenwood Ave.
Chicago, IL. 60637
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20140819/d50c9b20/attachment.htm