Hierarchical Multi-Task Transformers for Crosslingual Low Resource Phoneme Recognition
- This paper proposes a method for multilingual phoneme recognition in unseen, low resource languages. We propose a novel hierarchical multi-task classifier built on a hybrid convolution-transformer acoustic architecture where articulatory attribute and phoneme classifiers are optimized jointly. The model was evaluated on a subset of 24 languages from the Mozilla Common Voice corpus. We found that when using regular multi-task learning, negative transfer effects occurred between attribute and phoneme classifiers. They were reduced by the hierarchical architecture. When evaluating zero-shot crosslingual transfer on a data set with 95 languages, our hierarchical multi-task classifier achieves an absolute PER improvement of 2.78% compared to a phoneme-only baseline.
Author: | Kevin GlockerORCiD, Munir GeorgesORCiD |
---|---|
Language: | English |
Document Type: | Conference Paper |
Conference: | 5th International Conference on Natural Language and Speech Processing (ICNLSP 2022), Trento (Italy), 16.-17.12.2022 |
Year of first Publication: | 2022 |
published in (English): | Proceedings of the 5th International Conference on Natural Language and Speech Processing (ICNLSP 2022) |
Publisher: | Association for Computational Linguistics |
Place of publication: | Stroudsburg |
ISBN: | 978-1-959429-36-4 |
First Page: | 187 |
Last Page: | 192 |
Review: | peer-review |
Open Access: | ja |
Tag: | multi-task learning; multilingual; speech recognition; zero-shot |
URN: | urn:nbn:de:bvb:573-30544 |
URL: | https://aclanthology.org/2022.icnlsp-1.21 |
Faculties / Institutes / Organizations: | Fakultät Informatik |
AImotion Bavaria | |
Licence (German): | ![]() |
Release Date: | 2023/02/14 |