BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260522T150116Z
LOCATION:D167/174
DTSTART;TZID=America/Chicago:20181111T170000
DTEND;TZID=America/Chicago:20181111T173000
UID:submissions.supercomputing.org_SC18_sess221_ws_mlhpce122@linklings.com
SUMMARY:Training Speech Recognition Models on HPC Infrastructure
DESCRIPTION:Deepthi Karkada and Vikram A. Saletore (Intel Corporation)\n\n
 Automatic speech recognition is used extensively in speech interfaces and 
 spoken dialogue systems. To accelerate the development of new speech recog
 nition models and techniques, developers at Mozilla have open sourced a de
 ep learning based Speech-To-Text engine known as project DeepSpeech based 
 on Baidu’s DeepSpeech research. In order to make model training time quick
 er on CPUs for DeepSpeech distributed training, we have developed optimiza
 tions on the Mozilla DeepSpeech code to scale the model training to a larg
 e number of Intel® CPU system, including Horovod integration into DeepSpee
 ch. We have also implemented a novel dataset partitioning scheme to mitiga
 te compute imbalance across multiple nodes of an HPC cluster. We demonstra
 te that we are able to train the DeepSpeech model using the LibriSpeech cl
 ean dataset to its state-of-the-art accuracy in 6.45Hrs on 16-Node Intel® 
 Xeon® based HPC cluster.\n\nTag: Applications, Deep Learning, Machine Lear
 ning\n\nRegistration Category: Workshop Reg Pass\n\n
END:VEVENT
END:VCALENDAR