Feature Request: Add SenseVoice/Paraformer as STT engine option

[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.

Feature Request

Speech Note already supports multiple STT engines. Adding SenseVoice and Paraformer from FunASR would expand language coverage and improve performance.

Why

• SenseVoice — 234M param non-autoregressive ASR, 50+ languages, very fast inference
• Paraformer — excellent accuracy on Chinese, English, Japanese, Korean
• Both are available via Sherpa-ONNX which already supports Linux desktop and Flatpak — ideal for Speech Note's distribution model
• Built-in emotion detection and audio event classification
• Fully offline after model download

Integration Path

The easiest integration would be through Sherpa-ONNX (5K+ stars), which provides:

• C/C++ API suitable for dsnote's C++ codebase
• Pre-built ONNX models for SenseVoice and Paraformer
• Linux packages and Flatpak-friendly builds
• Already used in production by multiple Linux apps

Models

Model	Size	Languages	Speed
SenseVoiceSmall	234M	50+	170x real-time (GPU)
Paraformer-large	220M	zh/en/ja/ko/yue	160x real-time (GPU)

Happy to help with integration!