Feature: FunASR/SenseVoice STT extension as Whisper alternative

[LauraGPT]

Feature Request

The built-in whisper_stt extension works well but only supports Whisper models. A FunASR/SenseVoice STT extension would provide a faster, smaller alternative with richer output.

Why SenseVoice is compelling for text-generation-webui

	Whisper small.en	SenseVoice-Small
Params	244M	234M
Architecture	Autoregressive	Non-autoregressive
Languages	English only	50+ languages
Speed	Baseline	Up to 25x faster
Extras	—	Emotion + audio events

SenseVoice's non-autoregressive architecture gives constant-time decoding — faster response means more natural voice conversations.

Implementation

The extension structure would mirror whisper_stt, with FunASR as the backend:

from funasr import AutoModel

model = AutoModel(model="iic/SenseVoiceSmall")

def do_stt(audio):
    result = model.generate(input=audio_array)
    return result[0]["text"]

The funasr package installs via pip and supports both CPU and CUDA.

Happy to contribute a PR for this extension if there's interest.

References:

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