import base64
import os
import streamlit as st
from openai import OpenAI


def get_mimo_tts_client():
    """获取 mimo TTS 客户端（复用同一套密钥）"""
    api_key = os.getenv("MIMO_API_KEY")
    base_url = os.getenv("MIMO_BASE_URL")

    if not api_key:
        try:
            api_key = st.secrets["MIMO_API_KEY"]
        except Exception:
            pass

    if not base_url:
        try:
            base_url = st.secrets["MIMO_BASE_URL"]
        except Exception:
            base_url = "https://api.xiaomimimo.com/v1"

    if not api_key:
        return None

    return OpenAI(api_key=api_key, base_url=base_url)


# mimo-v2.5-tts 预置音色列表
PRESET_VOICES = {
    "冰糖": {"id": "冰糖", "lang": "中文", "gender": "女性"},
    "茉莉": {"id": "茉莉", "lang": "中文", "gender": "女性"},
    "苏打": {"id": "苏打", "lang": "中文", "gender": "男性"},
    "白桦": {"id": "白桦", "lang": "中文", "gender": "男性"},
    "Mia": {"id": "Mia", "lang": "英文", "gender": "女性"},
    "Chloe": {"id": "Chloe", "lang": "英文", "gender": "女性"},
    "Milo": {"id": "Milo", "lang": "英文", "gender": "男性"},
    "Dean": {"id": "Dean", "lang": "英文", "gender": "男性"},
}


def text_to_speech(
    text: str,
    voice: str = "冰糖",
    style: str = "",
) -> bytes:
    """调用 mimo-v2.5-tts 模型生成语音

    Args:
        text: 要合成的文本（放在 assistant 消息中）
        voice: 预置音色名称
        style: 风格指令（放在 user 消息中，可选）

    Returns:
        WAV 格式的音频字节数据
    """
    client = get_mimo_tts_client()
    if not client:
        raise RuntimeError("未配置 MIMO API Key")

    messages = []

    # user 消息：风格指令（可选）
    if style:
        messages.append({"role": "user", "content": style})
    else:
        messages.append({"role": "user", "content": ""})

    # assistant 消息：要合成的目标文本
    messages.append({"role": "assistant", "content": text})

    completion = client.chat.completions.create(
        model="mimo-v2.5-tts",
        messages=messages,
        audio={"format": "wav", "voice": voice},
    )

    message = completion.choices[0].message
    audio_bytes = base64.b64decode(message.audio.data)
    return audio_bytes