Model
Conformer-CTC-Large model for Spanish Automatic Speech Recognition, Trained on NeMo ASRSET 2.0
Use the NGC CLI to download:
Copied!
1 Version
1.8.0Selected03/02/2022 2:29 AM UTC1.46 GB Copied!
1.8.0Selected
03/02/2022 2:29 AM UTC1.46 GB
Copied!
Accuracy (beam search + LM, size=128, alpha=0.8, beta=1.5)
| Key | Value |
|---|---|
| MCV7.0 dev | 5.0 % |
| MLS test | 3.6 % |
| CALLHOME train | 20.7 % |
| CALLHOME devtest | 21.6 % |
| MLS dev | 3.6 % |
| CALLHOME evltest | 23.3 % |
| Voxpopuli dev | 5.5 % |
| MCV7.0 test | 5.5 % |
| Fisher test | 17.5 % |
| Voxpopuli test | 6.7 % |
| Fisher dev | 17.5 % |
Accuracy (greedy)
| Key | Value |
|---|---|
| MCV7.0 dev | 6.3 % |
| MLS test | 4.2 % |
| CALLHOME train | 22.4 % |
| CALLHOME devtest | 23.7 % |
| MLS dev | 4.3 % |
| CALLHOME evltest | 25.3 % |
| Voxpopuli dev | 6.1 % |
| MCV7.0 test | 6.9 % |
| Fisher test | 18.5 % |
| Voxpopuli test | 7.5 % |
| Fisher dev | 18.3 % |
Model
| Key | Value |
|---|---|
| Architecture | Conformer CTC |
| Outputs | Transcribed speech |
| Inputs | 16 kHz Mono-channel |