NVIDIA
Use the NGC CLI to download:
Copied!
Model
NVIDIA
STT Es Citrinet 1024 Gamma 0.25Citrinet 1024 with kernel scaling factor (gamma) of 25% trained on ASR Set 2.0 dataset
Use the NGC CLI to download:
Copied!
1 Version
1.8.0Selected03/02/2022 2:26 AM UTC1.48 GBCopied!
1.8.0Selected
03/02/2022 2:26 AM UTC1.48 GB
Copied!
Accuracy (beam search + LM, size=128, alpha=0.3, beta=1)
| Key | Value |
|---|---|
| MCV7.0 dev | 5.0 % |
| MLS test | 3.6 % |
| CALLHOME train | 18.1 % |
| CALLHOME devtest | 19.3 % |
| MLS dev | 3.2 % |
| CALLHOME evltest | 20.3 % |
| Voxpopuli dev | 5.2 % |
| MCV7.0 test | 5.7 % |
| Fisher test | 15.5 % |
| Voxpopuli test | 6.4 % |
| Fisher dev | 15.4 % |
Accuracy (greedy)
| Key | Value |
|---|---|
| MCV7.0 dev | 6.1 % |
| MLS test | 4.1 % |
| CALLHOME train | 19.1 % |
| CALLHOME devtest | 19.9 % |
| MLS dev | 3.5 % |
| CALLHOME evltest | 21.3 % |
| Voxpopuli dev | 5.6 % |
| MCV7.0 test | 6.8 % |
| Fisher test | 15.9 % |
| Voxpopuli test | 7.0 % |
| Fisher dev | 15.8 % |
Model
| Key | Value |
|---|---|
| Architecture | Citrinet |
| Outputs | Transcribed speech |
| Inputs | 16 kHz Mono-channel |