Sorry, you need to enable JavaScript to visit this website.

ADVANCING RNN TRANSDUCER TECHNOLOGY FOR SPEECH RECOGNITION

Citation Author(s):
George Saon, Zoltan Tueske, Daniel Bolanos and Brian Kingsbury
Submitted by:
George Saon
Last updated:
22 June 2021 - 9:32am
Document Type:
Presentation Slides
Document Year:
2021
Event:
Presenters Name:
George Saon
Paper Code:
SPE-2.1

Abstract 

Abstract: 

We investigate a set of techniques for RNN Transducers (RNN-Ts) that were instrumental in lowering the word error rate on three different tasks (Switchboard 300 hours, conversational Spanish 780 hours and conversational Italian 900 hours). The techniques pertain to architectural changes, speaker adaptation, language model fusion, model combination and general training recipe. First, we introduce a novel multiplicative integration of the encoder and prediction network vectors in the joint network (as opposed to additive). Second, we discuss the applicability of i-vector speaker adaptation to RNNTs in conjunction with data perturbation. Third, we explore the effectiveness of the recently proposed density ratio language model fusion for these tasks. Last but not least, we describe the other components of our training recipe and their effect on recognition performance. We report a 5.9% and 12.5% word error rate on the Switchboard and CallHome test sets of the NIST Hub5 2000 evaluation and a 12.7% WER on the Mozilla CommonVoice Italian test set.

up
0 users have voted:

Dataset Files

icassp2021-slides.pdf

(49)