#!/bin/bash
# 小米 MiMo TTS 封装脚本 — Hermes command provider 调用
# 输出: Opus格式音频文件（飞书语音条兼容）

INPUT="$1"
OUTPUT="$2"
VOICE="${3:-冰糖}"
KEY="tp-c374efhzmz9npodjz5wj1cm008lg4czce6v170nqxwemp12v"
API="https://token-plan-cn.xiaomimimo.com/v1/chat/completions"

TEXT=$(cat "$INPUT")
TMPWAV="/tmp/mimo_tts_$$.wav"
TMP_OPUS="/tmp/mimo_tts_$$.opus"

# 调用小米TTS API
RESP=$(curl -s -X POST "$API" \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d "$(python3 -c "
import json, sys
payload = {
    'model': 'mimo-v2.5-tts',
    'messages': [
        {'role': 'user', 'content': '请用温柔自然的语气，正常语速朗读'},
        {'role': 'assistant', 'content': sys.argv[1]}
    ],
    'audio': {'format': 'wav', 'voice': sys.argv[2]}
}
print(json.dumps(payload, ensure_ascii=False))
" "$TEXT" "$VOICE")")

# 解析base64音频到临时wav文件
echo "$RESP" | python3 -c "
import json, base64, sys
try:
    d = json.load(sys.stdin)
    audio = d['choices'][0]['message']['audio']['data']
    with open('$TMPWAV', 'wb') as f:
        f.write(base64.b64decode(audio))
except Exception as e:
    print(f'错误: {e}', file=sys.stderr)
    sys.exit(1)
"

if [ ! -f "$TMPWAV" ]; then
    echo "错误: WAV文件未生成" >&2
    exit 1
fi

# 转换为Opus格式
if ! python3 ~/.hermes/scripts/to_opus.py "$TMPWAV" "$TMP_OPUS" 2>&1; then
    rm -f "$TMPWAV"
    echo "错误: Opus转码失败" >&2
    exit 1
fi

# 覆盖Hermes期望的输出路径（即使扩展名是.wav，内容是opus）
mv "$TMP_OPUS" "$OUTPUT"
rm -f "$TMPWAV"

exit 0