Automatic Speech Recognition on CSJ
0.17Eval1 Error RateWhisper-large-v3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Whisper-large-v3Model params.=1,541M, Beam size=52023.12 | 0.17 | 0.156 | 0.156 | 0.46 | |
| Whisper-large-v3Model params.=1,541M, Beam size=12023.12 | 0.172 | 0.156 | 0.148 | 0.3 | |
| Whisper-large-v2Model params.=1,541M, Beam size=52023.12 | 0.179 | 0.163 | 0.163 | 0.49 | |
| Whisper-mediumModel params.=769M, Beam size=52023.12 | 0.185 | 0.173 | 0.149 | 0.28 | |
| Whisper-mediumModel params.=769M, Beam size=12023.12 | 0.193 | 0.176 | 0.15 | 0.21 | |
| Whisper-smallModel params.=244M, Beam size=52023.12 | 0.209 | 0.202 | 0.174 | 0.14 | |
| reazonspeech-espnet-v1Model params.=90M, Beam size=52023.12 | 0.214 | 0.186 | 0.153 | 0.2 | |
| Whisper-smallModel params.=244M, Beam size=12023.12 | 0.217 | 0.215 | 0.18 | 0.1 | |
| reazonspeech-espnet-v1Model params.=90M, Beam size=202023.12 | 0.218 | 0.191 | 0.155 | 0.27 | |
| reazonspeech-espnet-v1Model params.=90M, Beam size=12023.12 | 0.229 | 0.201 | 0.153 | 0.06 | |
| Whisper-large-v2Model params.=1,541M, Beam size=12023.12 | 0.237 | 0.171 | 0.196 | 0.31 | |
| Whisper-baseModel params.=74M, Beam size=52023.12 | 0.28 | 0.257 | 0.242 | 0.08 | |
| Whisper-baseModel params.=74M, Beam size=12023.12 | 0.299 | 0.275 | 0.259 | 0.06 | |
| Nue-ASRModel params.=3,708M, Beam size=1, DeepSpeed-Inference=true2023.12 | 0.309 | 0.266 | 0.229 | 0.15 | |
| Nue-ASRModel params.=3,708M, Beam size=1, DeepSpeed-Inference=false2023.12 | 0.31 | 0.266 | 0.229 | 0.22 |