Info: SenseVoice.cpp — similar pure C/C++ ASR, non-autoregressive architecture

Hi Salvatore! Love the approach of pure C inference for speech-to-text.

In case you're interested, there's a similar project for a different ASR model:

**[SenseVoice.cpp](https://github.com/lovemefan/SenseVoice.cpp)** — Pure C/C++ implementation of SenseVoice (550 stars)

## Key architectural difference

SenseVoice uses a **non-autoregressive** architecture:
- Single forward pass → entire transcription at once
- No sequential token generation → deterministic, constant time
- 234M params, 50+ languages

This means the C/C++ implementation is simpler (no autoregressive loop, no KV cache management) while being ~5x faster than Whisper-class models.

There's also **[FunASR-GGML](https://github.com/huaxin0/FunASR-GGML)** which takes the GGML approach (similar to whisper.cpp) for SenseVoice inference.

Just sharing in case there's interesting cross-pollination between these C-based ASR implementations. The non-autoregressive architecture makes for a very different (and arguably simpler) inference loop.

## Links
- SenseVoice model: https://github.com/FunAudioLLM/SenseVoice (8.3K stars)
- FunASR toolkit: https://github.com/modelscope/FunASR (16.7K stars)