Slovami AI models
Model files for Slovami AI, offline Android speech recognition and speaker diarization.
These are redistributed third-party models, not newly trained weights. Original copyright and licence terms remain in force.
| Model | Original project | Licence |
|---|---|---|
| GigaAM-v3 Russian ASR | https://github.com/salute-developers/GigaAM | MIT |
| Whisper English ASR | https://github.com/openai/whisper | MIT |
| CAM++ and ERes2Net speaker embeddings | https://github.com/modelscope/3D-Speaker | Apache-2.0 |
| Navec word embeddings | https://github.com/natasha/navec | MIT |
| Slovnet NER | https://github.com/natasha/slovnet | MIT |
Files retain original bytes. The manifest includes sizes and SHA-256 checksums.
The optional QVikhr model is not included.
The Slovami AI application downloads these files from this public repository and executes inference on the device.