sherpa-onnx streaming transducer export of nvidia/nemotron-3.5-asr-streaming-0.6b at a 1120 ms chunk size, exported with the official k2-fsa/sherpa-onnx export_onnx.py.

Unlike the int8-only upstream package, this repo also ships the fp32 graphs (encoder.onnx + encoder.data, decoder.onnx, joiner.onnx) for best accuracy, alongside the *.int8.onnx variants.

The encoder is a cache-aware streaming FastConformer with a 6th prompt_index input; the encoder metadata carries prompt_dictionary (100+ languages → prompt id) and auto_prompt_id (101 = whole-utterance auto-detect). Used by WinSTT.

Multilingual (100+ languages). License CC-BY-4.0 (inherited from the base model).