Speech Recognition on Multilingual LibriSpeech (MLS) (test)
0.0322WERgranite-speech-4.0-1b
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=Spanish2026.03 | 0.0322 | — | — | — | — | — | — | — | 779 | 3 | 50 | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=Spanish2026.03 | 0.0328 | — | — | — | — | — | — | — | 627 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=Spanish2026.03 | 0.0366 | — | — | — | — | — | — | — | 2,490 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=German2026.03 | 0.0442 | — | — | — | — | — | — | — | 755 | 3 | 48 | |
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=French2026.03 | 0.0448 | — | — | — | — | — | — | — | 658 | 1 | 39 | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=French2026.03 | 0.0451 | — | — | — | — | — | — | — | 634 | — | — | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=German2026.03 | 0.0459 | — | — | — | — | — | — | — | 663 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=English2026.03 | 0.0472 | — | — | — | — | — | — | — | 772 | 1 | 42 | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=English2026.03 | 0.0474 | — | — | — | — | — | — | — | 702 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=German2026.03 | 0.0491 | — | — | — | — | — | — | — | 2,755 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=Average2026.03 | 0.0533 | — | — | — | — | — | — | — | 699 | 3 | 38 | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=Average2026.03 | 0.0547 | — | — | — | — | — | — | — | 629 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=French2026.03 | 0.0552 | — | — | — | — | — | — | — | 2,834 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=Average2026.03 | 0.0573 | — | — | — | — | — | — | — | 2,753 | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=English2026.03 | 0.0575 | — | — | — | — | — | — | — | 2,983 | — | — | |
| Whispermode=zero-shot2022.12 | 0.073 | — | — | — | — | — | — | — | — | — | — | |
| granite-speech-4.0-1bDecoding Regime=High RTFx SSD, tau_CTC=3.0, tau_SLM=0.1, Target Language=Portuguese2026.03 | 0.0883 | — | — | — | — | — | — | — | 2,701 | — | — | |
| mSLAM-CTCnumber of parameters=2B2022.12 | 0.097 | — | — | — | — | — | — | — | — | — | — | |
| granite-speech-4.0-1bDecoding Regime=High Accuracy SSD, tau_CTC=0.7, tau_SLM=0.2, Target Language=Portuguese2026.03 | 0.0984 | — | — | — | — | — | — | — | 531 | 8 | 10 | |
| granite-speech-4.0-1bDecoding Regime=Full AR, Target Language=Portuguese2026.03 | 0.1025 | — | — | — | — | — | — | — | 521 | — | — | |
| XLS-Rnumber of parameters=1B2022.12 | 0.109 | — | — | — | — | — | — | — | — | — | — | |
| REBORN2024.02 | — | 0.209 | 0.269 | 0.282 | 0.247 | 0.399 | 0.515 | 0.32 | — | — | — | |
| wav2vec-Usource=Baevski et al. [4]2024.02 | — | 0.325 | 0.402 | 0.398 | 0.333 | 0.581 | 0.598 | 0.44 | — | — | — | |
| wav2vec-Usource=Our reproduction2024.02 | — | 0.339 | 0.381 | 0.377 | 0.331 | 0.518 | 0.594 | 0.423 | — | — | — | |
| wav2vec-U 2.0source=Liu et al. [30]2024.02 | — | 0.235 | 0.351 | 0.357 | 0.258 | 0.469 | 0.485 | 0.359 | — | — | — |