Towards scalable efficient on-device ASR with transfer learning
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866929433039863808 |
|---|---|
| author | Pandey, Laxmi Li, Ke Guo, Jinxi Paul, Debjyoti Guo, Arthur Mahadeokar, Jay Zhang, Xuedong |
| author_facet | Pandey, Laxmi Li, Ke Guo, Jinxi Paul, Debjyoti Guo, Arthur Mahadeokar, Jay Zhang, Xuedong |
| contents | Multilingual pretraining for transfer learning significantly boosts the robustness of low-resource monolingual ASR models. This study systematically investigates three main aspects: (a) the impact of transfer learning on model performance during initial training or fine-tuning, (b) the influence of transfer learning across dataset domains and languages, and (c) the effect on rare-word recognition compared to non-rare words. Our finding suggests that RNNT-loss pretraining, followed by monolingual fine-tuning with Minimum Word Error Rate (MinWER) loss, consistently reduces Word Error Rates (WER) across languages like Italian and French. WER Reductions (WERR) reach 36.2% and 42.8% compared to monolingual baselines for MLS and in-house datasets. Out-of-domain pretraining leads to 28% higher WERR than in-domain pretraining. Both rare and non-rare words benefit, with rare words showing greater improvements with out-of-domain pretraining, and non-rare words with in-domain pretraining. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_16664 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Towards scalable efficient on-device ASR with transfer learning Pandey, Laxmi Li, Ke Guo, Jinxi Paul, Debjyoti Guo, Arthur Mahadeokar, Jay Zhang, Xuedong Computation and Language Audio and Speech Processing Multilingual pretraining for transfer learning significantly boosts the robustness of low-resource monolingual ASR models. This study systematically investigates three main aspects: (a) the impact of transfer learning on model performance during initial training or fine-tuning, (b) the influence of transfer learning across dataset domains and languages, and (c) the effect on rare-word recognition compared to non-rare words. Our finding suggests that RNNT-loss pretraining, followed by monolingual fine-tuning with Minimum Word Error Rate (MinWER) loss, consistently reduces Word Error Rates (WER) across languages like Italian and French. WER Reductions (WERR) reach 36.2% and 42.8% compared to monolingual baselines for MLS and in-house datasets. Out-of-domain pretraining leads to 28% higher WERR than in-domain pretraining. Both rare and non-rare words benefit, with rare words showing greater improvements with out-of-domain pretraining, and non-rare words with in-domain pretraining. |
| title | Towards scalable efficient on-device ASR with transfer learning |
| topic | Computation and Language Audio and Speech Processing |
| url | https://arxiv.org/abs/2407.16664 |