Feature: FunASR/SenseVoice as STT option — fast, multilingual, small model

[LauraGPT]

Feature Request

ElatoAI supports 100+ models for AI devices. Adding FunASR/SenseVoice as a STT option would give users a fast, compact alternative for voice recognition:

• SenseVoice-Small: 234M params, 50+ languages, non-autoregressive (constant-time decoding)
• Up to 25x faster than Whisper-large — important for real-time device interactions
• Built-in extras: Emotion detection + audio event classification at no extra cost
• OpenAI-compatible API: funasr-server exposes /v1/audio/transcriptions
• Edge options: Available via Sherpa-ONNX for on-device inference (iOS, Android, Raspberry Pi)

The Sherpa-ONNX path could be especially relevant for ESP32-connected devices where a lightweight ONNX model runs on the edge or a nearby server.

References:

• FunASR: https://github.com/modelscope/FunASR (16K+ stars)
• Sherpa-ONNX: https://github.com/k2-fsa/sherpa-onnx (edge deployment)