建议集成 FunASR/SenseVoice 替代实时语音对话的 ASR 模块

## 功能建议

gpt_academic 已有实时语音对话功能，非常出色！建议集成 FunASR/SenseVoice 作为 ASR 引擎选项。

**为什么选择 FunASR/SenseVoice？**

- **SenseVoice**：超快非自回归 ASR（234M 参数），中文识别精度极高
- **比 Whisper 快 5-10 倍**：非自回归架构，更适合实时对话场景
- **情感检测**：内置识别语音情感
- **音频事件检测**：识别笑声、掌声等
- **Fun-ASR-Nano**：基于 LLM 的 ASR（800M 参数），支持热词定制
- **流式识别**：支持 WebSocket 实时流式 ASR
- **OpenAI 兼容 API**：`funasr-server` 提供标准接口

**快速接入：**
```python
from funasr import AutoModel
model = AutoModel(model="iic/SenseVoiceSmall")
result = model.generate(input="audio.wav")
```

**或通过 API 服务：**
```bash
funasr-server --device cuda
# /v1/audio/transcriptions 接口
```

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

Start by locating the existing real-time voice conversation ASR integration and identify how an engine option is configured. Compare the proposed FunASR AutoModel path with the funasr-server `/v1/audio/transcriptions` API and its WebSocket streaming support. Done should mean FunASR/SenseVoice can be selected as an ASR option for real-time conversations.

Written by the indexing model from the issue text.