Feature: Add SenseVoice/Paraformer ASR models for on-device speech recognition

[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

Cactus focuses on low-latency AI for mobile devices. SenseVoice and Paraformer are excellent candidates for on-device speech recognition:

Why SenseVoice fits Cactus:

• Non-autoregressive — constant-time inference regardless of audio length
• Small model — 234M parameters (SenseVoice-Small)
• ONNX ready — pre-built ONNX models available via Sherpa-ONNX
• 50+ languages — single model for multilingual recognition
• Extra features — emotion detection, audio event detection (laughter, applause, music)
• Mobile-proven — Sherpa-ONNX runs SenseVoice on iOS, Android, and WASM

Pre-built ONNX models and C/C++ bindings:

• https://github.com/k2-fsa/sherpa-onnx

References:

• FunASR — 16K+ GitHub stars
• SenseVoice — 8K+ stars