IndicConformer Punjabi — sherpa-onnx (mirror)
Punjabi (Gurmukhi) on-device speech recognition via sherpa-onnx.
Mirror of the pa export from
parismitaglobalsolutions/indicconformer-sherpa-onnx.
Contents
tokens.txt— shared IndicConformer tokenizer vocabulary (67,605 bytes)model.int8.onnx— Punjabi Conformer CTC model, INT8-quantized ONNX (~188 MB)
Usage (sherpa-onnx)
OfflineNemoEncDecCtcModelConfig nemo = new OfflineNemoEncDecCtcModelConfig();
nemo.setModel("model.int8.onnx");
modelConfig.setNemo(nemo);
modelConfig.setTokens("tokens.txt");
config.setDecodingMethod("greedy_search");
Credits & License
- Model weights & architecture: IndicConformer by AI4Bharat (IIT Madras) — MIT (source repo)
- ONNX export & INT8 quantization: parismitaglobalsolutions — Apache-2.0