Nemotron-3.5 ASR Streaming 0.6B — Multilingual ONNX
Pre-exported ONNX files from nvidia/nemotron-3.5-asr-streaming-0.6b for use with sherpa-onnx and parakeet-rs. Supports 40 language-locales with streaming inference and auto language detection.
Source: altunenes/parakeet-rs
Files
| File | Size | Description |
|---|---|---|
encoder.onnx + encoder.onnx.data |
~2.4 GB | Streaming encoder with prompt_index input |
decoder_joint.onnx |
93 MB | Decoder + joint network |
tokenizer.model |
0.4 MB | SentencePiece tokenizer (13k vocab) |
Encoder Inputs
| Input | Shape | Description |
|---|---|---|
processed_signal |
[batch, 128, time] | Mel spectrogram |
processed_signal_length |
[batch] | Signal length |
cache_last_channel |
[24, 1, 56, 1024] | Channel cache |
cache_last_time |
[24, 1, 1024, 8] | Time cache |
cache_last_channel_len |
[1] | Cache length |
prompt_index |
[batch] | Language index (101=auto) |
Usage (sherpa-onnx)
Configure as nemo_transducer model type with OnlineRecognizer for streaming inference. Pass prompt_index=101 for auto language detection or a specific language index.
License
NVIDIA Open Model License — see LICENSE.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for pantinor/nemotron-3.5-asr-streaming-0.6b-onnx
Base model
nvidia/nemotron-3.5-asr-streaming-0.6b