A Khalis Foundation board for Punjabi Gurbani speech recognition models.
Scores are not published yet. The metric columns stay empty until a shared evaluation is agreed and run. No training data, evaluation set, or score is listed on this page.
| Model | Architecture | License | Size | WER | CER | nCER |
|---|---|---|---|---|---|---|
| BaniGuru v1 | IndicConformer hybrid CTC-RNNT | MIT | 523 MB | — | — | — |
| IndicConformer Punjabi v3 Kirtan | IndicConformer hybrid CTC-RNNT | Apache-2.0 | 138 MB | — | — | — |
| IndicConformer Punjabi CTC shabad preview | IndicConformer CTC (ONNX) | MIT | 184 MB | — | — | — |
Size is the smallest published file that can run on its own, not the parameter count. All three are about 120M parameters. Models are listed in the order they were added, not by rank. A dash means the score has not been published.