OpenClaw TTS语音合成优化方案：小米MiMo直连方案#

1
brew install ffmpeg

1
{
2
  "messages": {
3
    "tts": {
4
      "auto": "always",
5
      "provider": "xiaomi-mimo",
6
      "providers": {
7
        "xiaomi-mimo": {
8
          "model": "mimo-v2-tts",
9
          "voice": "default_zh"
10
        }
11
      }
12
    }
13
  }
14
}

1
#!/usr/bin/env python3
2
import requests, base64, subprocess, os
3

4
API_URL = "https://api.xiaomimimo.com/v1/chat/completions"
5
MIMO_API_KEY = os.environ.get("MIMO_API_KEY", "")
6

7
def tts_send(text, output_path):
8
    # 1. 调用MiMo TTS API
9
    resp = requests.post(API_URL,
10
        headers={
11
            "api-key": MIMO_API_KEY,
12
            "Content-Type": "application/json"
13
        },
14
        json={
15
            "model": "mimo-v2-tts",
16
            "messages": [{"role": "assistant", "content": text}],
17
            "audio": {"format": "mp3", "voice": "default_zh"}
18
        },
19
        timeout=30)
20

21
    # 2. 解码音频
22
    audio_data = resp.json()["choices"][0]["message"]["audio"]["data"]
23
    mp3_bytes = base64.b64decode(audio_data)
24

25
    # 3. 保存并转opus
26
    mp3_path = "/tmp/tts.mp3"
27
    with open(mp3_path, "wb") as f:
28
        f.write(mp3_bytes)
29

30
    subprocess.run([
31
        "ffmpeg", "-i", mp3_path,
32
        "-acodec", "libopus", "-ac", "1", "-ar", "16000",
33
        output_path, "-y"
34
    ], capture_output=True)
35

36
    os.unlink(mp3_path)
37
    return output_path

1
# 合成语音
2
opus_path = tts_send("臣妾在呢～", "/tmp/voice.opus")
3

4
# 通过qqmedia发送
5
# <qqmedia>/path/to/voice.opus</qqmedia>