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

facebooktwittermailshare

Learning from the best: A teacher-student multilingual framework for low-resource languages

Abstract: 

The traditional method of pretraining neural acoustic models in low-resource languages consists of initializing the acoustic model parameters with a large, annotated multilingual corpus and can be a drain on time and resources. In an attempt to reuse TDNN-LSTMs already pre-trained using multilingual training, we have applied Teacher-Student (TS) learning as a method of pretraining to transfer knowledge from a multilingual TDNN-LSTM to a TDNN. The pretraining time is reduced by an order of magnitude with the use of language-specific data during the teacher-student training. Additionally, the TS architecture allows us to leverage untranscribed data, previously untouched during supervised training. The best student TDNN achieves a WER within 1% of the teacher TDNN-LSTM performance and shows consistent improvement in recognition over TDNNs trained using the traditional pipeline over all the evaluation languages. Switching to TDNN from TDNN-LSTM also allows sub-real time decoding.

up
0 users have voted:

Paper Details

Authors:
Deblin Bagchi and William Hartmann
Submitted On:
13 May 2019 - 5:43pm
Short Link:
Type:
Poster
Event:
Presenter's Name:
Deblin Bagchi
Paper Code:
SLP-P3.1
Document Year:
2019
Cite

Document Files

ICASSP_2019_poster_multi_deblin_bagchi

(5)

Subscribe

[1] Deblin Bagchi and William Hartmann, "Learning from the best: A teacher-student multilingual framework for low-resource languages", IEEE SigPort, 2019. [Online]. Available: http://sigport.org/4493. Accessed: May. 22, 2019.
@article{4493-19,
url = {http://sigport.org/4493},
author = {Deblin Bagchi and William Hartmann },
publisher = {IEEE SigPort},
title = {Learning from the best: A teacher-student multilingual framework for low-resource languages},
year = {2019} }
TY - EJOUR
T1 - Learning from the best: A teacher-student multilingual framework for low-resource languages
AU - Deblin Bagchi and William Hartmann
PY - 2019
PB - IEEE SigPort
UR - http://sigport.org/4493
ER -
Deblin Bagchi and William Hartmann. (2019). Learning from the best: A teacher-student multilingual framework for low-resource languages. IEEE SigPort. http://sigport.org/4493
Deblin Bagchi and William Hartmann, 2019. Learning from the best: A teacher-student multilingual framework for low-resource languages. Available at: http://sigport.org/4493.
Deblin Bagchi and William Hartmann. (2019). "Learning from the best: A teacher-student multilingual framework for low-resource languages." Web.
1. Deblin Bagchi and William Hartmann. Learning from the best: A teacher-student multilingual framework for low-resource languages [Internet]. IEEE SigPort; 2019. Available from : http://sigport.org/4493