navi-org/navi-voice-nemotron-3.5-asr-streaming-0.6b-onnx

NAVI Voice — Nemotron 3.5 ASR Streaming 0.6B (ONNX INT4)

Packaged weights for local dictation in NAVI.

• —Engine id: nemotron_streaming
• —Runtime: ONNX Runtime (ort in Rust)
• —Sample rate: 16 kHz mono
• —Chunk: 8960 samples (~560 ms) per genai_config.json
• —Streaming: yes (cache-aware FastConformer-RNNT)
• —Size on disk: ~757 MB

Install with NAVI

navi voice init --engine nemotron_streaming
navi voice doctor

Models are stored under {data_dir}/voice/models/ (not inside project worktrees).

Layout

navi-manifest.json          # NAVI install index
LICENSE                     # OpenMDW-1.1
NOTICE                      # Origin / attribution
SHA256SUMS                  # Integrity for init/doctor
genai_config.json           # Streaming graph + I/O names
audio_processor_config.json # Feature extraction params
tokenizer.json / vocab.txt
onnx/
  encoder.onnx[.data]
  decoder.onnx[.data]
  joint.onnx[.data]
  silero_vad.onnx

License

OpenMDW-1.1 — see:

• —LICENSE
• —NOTICE

Base model: `nvidia/nemotron-3.5-asr-streaming-0.6b`. ONNX INT4 packaging derived from `onnx-community/nemotron-3.5-asr-streaming-0.6b-onnx-int4`.

If you redistribute these files, retain this license and origin notices.

Notes

• —This is a redistribution / packaging for NAVI offline install, not a re-train.
• —Outputs (transcriptions) are not restricted by OpenMDW beyond the license text.
• —Prefer SHA256SUMS verification after download.