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
|
Download ANDROID_VALIDATION.md from spybyscript/Nemotron-3-Diarization-litert: direct link, hf CLI and curl.
- Browser
- Download file 1.13 kB
-
https://huggingface.co/spybyscript/Nemotron-3-Diarization-litert/resolve/main/ANDROID_VALIDATION.md
- Command line
-
hf download hf://spybyscript/Nemotron-3-Diarization-litert/ANDROID_VALIDATION.md
-
curl -L -o ANDROID_VALIDATION.md https://huggingface.co/spybyscript/Nemotron-3-Diarization-litert/resolve/main/ANDROID_VALIDATION.md
1.13 kB
| # 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. | |