Feature Proposal: Add FunASR as Recommended Production Inference Backend

ESPnet is an excellent end-to-end ASR toolkit. However, for production deployments requiring fast inference, FunASR (17.8K+ stars, https://github.com/modelscope/FunASR) offers complementary capabilities:

• SenseVoice: Ultra-fast multilingual ASR (50x faster than Whisper-large) — useful for real-time applications
• Paraformer: Non-autoregressive ASR with timestamps and punctuation — fast parallel inference
• OpenAI-compatible API: POST /v1/audio/transcriptions — easy deployment

For ESPnet users who need production-ready, high-speed ASR alongside ESPnets research-oriented training pipeline, FunASR provides a deployment-friendly alternative. The two toolkits complement each other — ESPnet for research/training, FunASR for production inference.

Would it be useful to add FunASR as a recommended inference backend in ESPnets documentation?