Feature Request: Add SenseVoice model for faster on-device STT

[LauraGPT]

Feature Request

Box is a great private on-device AI suite for Android. Suggesting SenseVoice as an STT model option.

Why SenseVoice for Android on-device?

• 234M params — lightweight for mobile, fast model loading
• Non-autoregressive — constant low latency, ideal for interactive use
• 5x faster than Whisper — less battery drain
• 50+ languages — single model for any locale
• Emotion detection — unique for mobile AI
• Already on Android — available via Sherpa-ONNX Android SDK

Resources

• FunASR: https://github.com/modelscope/FunASR (16.6K stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8.3K stars)
• Sherpa-ONNX Android: https://github.com/k2-fsa/sherpa-onnx

[jegly]

Thank you this looks promising ! @LauraGPT

[jegly]

@LauraGPT its coming !

[jegly]

@LauraGPT it has been implemented, and it is very fast, it will be in the next release i will ping you once its dropped !

[jegly]

ping!