Gender domain adaptation for automatic speech recognition task
This paper is focused on the finetuning of acoustic models for speaker adaptation goals on a given gender. We pretrained the Transformer baseline model on Librispeech-960 and conduct experiments with finetuning on the gender-specific test subsets and. In general, we do not obtain essential WER reduc...
Main Authors: | , |
---|---|
Format: | Article in Journal/Newspaper |
Language: | unknown |
Published: |
arXiv
2020
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.2010.04224 https://arxiv.org/abs/2010.04224 |
id |
ftdatacite:10.48550/arxiv.2010.04224 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.2010.04224 2023-05-15T14:57:16+02:00 Gender domain adaptation for automatic speech recognition task Artem, Sokolov Savchenko, Andrey V. 2020 https://dx.doi.org/10.48550/arxiv.2010.04224 https://arxiv.org/abs/2010.04224 unknown arXiv arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ Audio and Speech Processing eess.AS Sound cs.SD FOS Electrical engineering, electronic engineering, information engineering FOS Computer and information sciences Article CreativeWork article Preprint 2020 ftdatacite https://doi.org/10.48550/arxiv.2010.04224 2022-03-10T15:15:57Z This paper is focused on the finetuning of acoustic models for speaker adaptation goals on a given gender. We pretrained the Transformer baseline model on Librispeech-960 and conduct experiments with finetuning on the gender-specific test subsets and. In general, we do not obtain essential WER reduction by finetuning techniques by this approach. We achieved up to ~5% lower word error rate on the male subset and 3% on the female subset if the layers in the encoder and decoder are not frozen, but the tuning is started from the last checkpoints. Moreover, we adapted our base model on the full L2 Arctic dataset of accented speech and fine-tuned it for particular speakers and male and female genders separately. The models trained on the gender subsets obtained 1-2% higher accuracy when compared to the model tuned on the whole L2 Arctic dataset. Finally, we tested the concatenation of the pretrained x-vector voice embeddings and embeddings from a conventional encoder, but its gain in accuracy is not significant. : Draft of paper for SAMI conference Article in Journal/Newspaper Arctic sami DataCite Metadata Store (German National Library of Science and Technology) Arctic |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
unknown |
topic |
Audio and Speech Processing eess.AS Sound cs.SD FOS Electrical engineering, electronic engineering, information engineering FOS Computer and information sciences |
spellingShingle |
Audio and Speech Processing eess.AS Sound cs.SD FOS Electrical engineering, electronic engineering, information engineering FOS Computer and information sciences Artem, Sokolov Savchenko, Andrey V. Gender domain adaptation for automatic speech recognition task |
topic_facet |
Audio and Speech Processing eess.AS Sound cs.SD FOS Electrical engineering, electronic engineering, information engineering FOS Computer and information sciences |
description |
This paper is focused on the finetuning of acoustic models for speaker adaptation goals on a given gender. We pretrained the Transformer baseline model on Librispeech-960 and conduct experiments with finetuning on the gender-specific test subsets and. In general, we do not obtain essential WER reduction by finetuning techniques by this approach. We achieved up to ~5% lower word error rate on the male subset and 3% on the female subset if the layers in the encoder and decoder are not frozen, but the tuning is started from the last checkpoints. Moreover, we adapted our base model on the full L2 Arctic dataset of accented speech and fine-tuned it for particular speakers and male and female genders separately. The models trained on the gender subsets obtained 1-2% higher accuracy when compared to the model tuned on the whole L2 Arctic dataset. Finally, we tested the concatenation of the pretrained x-vector voice embeddings and embeddings from a conventional encoder, but its gain in accuracy is not significant. : Draft of paper for SAMI conference |
format |
Article in Journal/Newspaper |
author |
Artem, Sokolov Savchenko, Andrey V. |
author_facet |
Artem, Sokolov Savchenko, Andrey V. |
author_sort |
Artem, Sokolov |
title |
Gender domain adaptation for automatic speech recognition task |
title_short |
Gender domain adaptation for automatic speech recognition task |
title_full |
Gender domain adaptation for automatic speech recognition task |
title_fullStr |
Gender domain adaptation for automatic speech recognition task |
title_full_unstemmed |
Gender domain adaptation for automatic speech recognition task |
title_sort |
gender domain adaptation for automatic speech recognition task |
publisher |
arXiv |
publishDate |
2020 |
url |
https://dx.doi.org/10.48550/arxiv.2010.04224 https://arxiv.org/abs/2010.04224 |
geographic |
Arctic |
geographic_facet |
Arctic |
genre |
Arctic sami |
genre_facet |
Arctic sami |
op_rights |
arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
op_doi |
https://doi.org/10.48550/arxiv.2010.04224 |
_version_ |
1766329346350383104 |