Speech-to-speech translation for a real-world unwritten language
We study speech-to-speech translation (S2ST) that translates speech from one language into another language and focuses on building systems to support languages without standard text writing systems. We use English-Taiwanese Hokkien as a case study, and present an end-to-end solution from training data collection, modeling choices to benchmark dataset release. First, we present efforts on creating human annotated data, automatically mining data from large unlabeled speech datasets, and adopting pseudo-labeling to produce weakly supervised data. On the modeling, we take advantage of recent advances in applying self-supervised discrete representations as target for prediction in S2ST and show the effectiveness of leveraging additional text supervision from Mandarin, a language similar to Hokkien, in model training. Finally, we release an S2ST benchmark set to facilitate future research in this field.
PDF AbstractCode
Datasets
LibriSpeech
MuST-C
WenetSpeech
TAT
Results from the Paper
Ranked #1 on
Speech-to-Speech Translation
on TAT
(using extra training data)
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
|---|---|---|---|---|---|---|---|
| Speech-to-Speech Translation | TAT | Hokkien→En (Two-pass decoding) | ASR-BLEU (Dev) | 13.6 | # 1 | ||
| ASR-BLEU (Test) | 12.5 | # 1 | |||||
| Speech-to-Speech Translation | TAT | En→Hokkien (Single-pass decoding) | ASR-BLEU (Dev) | 6.6 | # 8 | ||
| ASR-BLEU (Test) | 6.0 | # 8 | |||||
| Speech-to-Speech Translation | TAT | En→Hokkien (Two-stage) | ASR-BLEU (Dev) | 7.1 | # 7 | ||
| ASR-BLEU (Test) | 6.6 | # 7 | |||||
| Speech-to-Speech Translation | TAT | En→Hokkien (Three-stage) | ASR-BLEU (Dev) | 7.5 | # 6 | ||
| ASR-BLEU (Test) | 6.8 | # 6 | |||||
| Speech-to-Speech Translation | TAT | En→Hokkien (Two-pass decoding) | ASR-BLEU (Dev) | 7.8 | # 5 | ||
| ASR-BLEU (Test) | 7.3 | # 5 | |||||
| Speech-to-Speech Translation | TAT | Hokkien→En (Single-pass decoding) | ASR-BLEU (Dev) | 8.8 | # 4 | ||
| ASR-BLEU (Test) | 8.1 | # 4 | |||||
| Speech-to-Speech Translation | TAT | Hokkien→En (Three-stage) | ASR-BLEU (Dev) | 12.5 | # 2 | ||
| ASR-BLEU (Test) | 8.8 | # 3 | |||||
| Speech-to-Speech Translation | TAT | Hokkien→En (Two-stage) | ASR-BLEU (Dev) | 12.5 | # 2 | ||
| ASR-BLEU (Test) | 10.5 | # 2 |
