Comparison / interop with FunASR and SenseVoice

[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! Interesting multilingual ASR model.

Wanted to suggest potential interop or comparison with FunASR and SenseVoice:

FunASR ecosystem

• FunASR — Industrial-grade ASR toolkit with streaming, VAD, punctuation, diarization
• SenseVoice — Multi-task speech model (ASR + emotion + audio events), 50+ languages
• Fun-ASR-Nano — Encoder-LLM architecture, HuggingFace Transformers native

Potential collaboration areas

1. Benchmark comparison — Could be valuable to compare Dolphin with Paraformer/SenseVoice on multilingual benchmarks
2. Pipeline integration — FunASR's VAD (FSMN-VAD) and punctuation models could complement Dolphin
3. Streaming — FunASR's streaming architecture could pair with Dolphin's recognition

References

• FunASR: https://github.com/modelscope/FunASR (16K+ stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8K+ stars)
• Benchmarks: https://funasr.com