Instructions to use spybyscript/Nemotron-3-Diarization-litert with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LiteRT
How to use spybyscript/Nemotron-3-Diarization-litert with LiteRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
File size: 1,134 Bytes
47382f6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 | # Android validation
The fused v2 artifact was validated on a Samsung Galaxy S23 Ultra (`SM-S918U1`,
Android API 36) using the LiteRT CPU/XNNPACK path with four threads.
The harness fed a mono PCM16 16 kHz multilingual, three-speaker WAV directly to
the streaming API in 250 ms blocks. It did not use media playback or the device
microphone. The recording was 182.053 seconds / 5,825,786 bytes with SHA-256
`54ea7e35b21e7a8206c16bd7be2e94701dbca85ef5647862affcfa683a6cb36c`.
The audio is not redistributed with this model.
| Measurement | Result |
| --- | ---: |
| Status | PASS |
| Accepted audio | 182,053 ms |
| Elapsed wall time | 112,252 ms |
| Wall real-time factor | 0.6166 |
| First segment | 1,250 ms audio / 814 ms wall |
| Slowest 250 ms append | 735 ms |
| Loaded process PSS | 199,252 KiB |
| Peak process PSS | 352,057 KiB |
| Stabilized segments | 38 |
| Detected speaker channels | 3 |
This is one device run, not a guarantee for other devices. The audio has no
human reference annotation, so the run validates execution, streaming latency,
resource measurement, and output stability—not diarization error rate.
|