Use the NGC CLI to download:
Copied!
Model
This collection contains the large version (114M) of the Persian speech recognition model with a FastConformer encoder and a Hybrid decoder (joint RNNT-CTC loss). The model has a vocab size of 1024.
Use the NGC CLI to download:
Copied!
1 Version
1.21.0Selected11/07/2023 6:33 PM UTC437.96 MBCopied!
1.21.0Selected
11/07/2023 6:33 PM UTC437.96 MB
Copied!
Accuracy
| Key | Value |
|---|---|
| CTC WER CV-Fa dev (custom) | 13.18 |
| RNNT CER CV-Fa dev (custom) | 3.89 |
| CTC CER CV-Fa test (custom) | 3.85 |
| RNNT WER CV-Fa test (custom) | 15.48 |
| CTC WER CV-Fa test (custom) | 13.16 |
| RNNT WER CV-Fa dev (custom) | 15.44 |
| RNNT CER CV-Fa test (custom) | 4.63 |
| CTC CER CV-Fa dev (custom) | 3.38 |
Model
| Key | Value |
|---|---|
| Architecture | FastConformer Hybrid Transducer-CTC Large |
| Outputs | Sequence of BPE tokens representing recognized text |
| Inputs | Sequence of 16kHz single-channel audio samples |