pith. sign in

arxiv: 1812.03919 · v2 · pith:2HWI33V7new · submitted 2018-12-10 · 📡 eess.AS · cs.CL· cs.SD

Pretraining by Backtranslation for End-to-end ASR in Low-Resource Settings

classification 📡 eess.AS cs.CLcs.SD
keywords languagespretrainingrelativespeechtext-basedtranscribeddatalow-resource
0
0 comments X
read the original abstract

We explore training attention-based encoder-decoder ASR in low-resource settings. These models perform poorly when trained on small amounts of transcribed speech, in part because they depend on having sufficient target-side text to train the attention and decoder networks. In this paper we address this shortcoming by pretraining our network parameters using only text-based data and transcribed speech from other languages. We analyze the relative contributions of both sources of data. Across 3 test languages, our text-based approach resulted in a 20% average relative improvement over a text-based augmentation technique without pretraining. Using transcribed speech from nearby languages gives a further 20-30% relative reduction in character error rate.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.