Feature: Add SenseVoice for private local transcription

[LauraGPT]

Note

License and capability clarification (2026-07-14): FunASR is a toolkit, not a single checkpoint. The FunASR and SenseVoice repository source code is MIT; model weights follow each model card. SenseVoiceSmall supports Chinese, Cantonese, English, Japanese, and Korean, and its weights use the linked FunASR Model Open Source License Agreement. Fun-ASR-Nano-2512 is Apache-2.0. Language coverage, punctuation, and performance depend on the selected model and runtime configuration.

Hi! NotelyVoice looks great — 100% private AI voice transcription.

Since privacy is core to your app, SenseVoice aligns perfectly:

Why SenseVoice?

• 100% offline — no API calls, no data leaves the device
• 5x faster than Whisper — non-autoregressive
• 234M params — runs on CPU, light on battery
• Emotion detection — tag notes with detected mood
• 50+ languages — auto-detection

Integration

from funasr import AutoModel
model = AutoModel(model="iic/SenseVoiceSmall", vad_model="fsmn-vad", punc_model="ct-punc")
result = model.generate(input="recording.wav")

For native apps (no Python):

• Sherpa-ONNX: https://github.com/k2-fsa/sherpa-onnx (Swift/C++/Rust bindings)

Links

• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8.3K stars)
• FunASR: https://github.com/modelscope/FunASR (16.7K stars)

[tosinonikute]

Thanks @LauraGPT for the links & suggestion, I'm particularly excited as it comes with Speaker Diarization, I've been struggling to implement Diarization with Whisper.cpp. I've have tried using Sherpa-ONNX with Whisper.cpp before and It was extremely slow.

I'll try the Android demo and compare.

[LauraGPT]

\nImportant correction to my original suggestion: SenseVoiceSmall itself does **not** provide speaker diarization. In the FunASR toolkit, diarization requires a separate speaker model/pipeline such as CAM++ and timestamp/alignment logic; a sherpa-onnx Android SenseVoice integration would not automatically include that capability. SenseVoiceSmall's documented ASR languages are Mandarin Chinese, Cantonese, English, Japanese, and Korean.\n\nI am sorry the earlier title/body led you to expect built-in diarization. If diarization is a hard requirement, it should be evaluated as a separate integration alongside SenseVoice rather than attributed to the checkpoint.