app.py

import os, shutil, base64, uuid, mimetypes
from pydub import AudioSegment
from openai import OpenAI
import gradio as gr

# ====== 基本設定 ======
PASSWORD = os.getenv("APP_PASSWORD", "chou")
MAX_SIZE = 25 * 1024 * 1024
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

print("===== 🚀 啟動中 =====")
print(f"APP_PASSWORD: {'✅ 已載入' if PASSWORD else '❌ 未載入'}")

# ====== 工具：把 data:URL 轉成臨時檔 ======
MIME_EXT = {
    "audio/mp4": "m4a",
    "audio/m4a": "m4a",
    "audio/aac": "aac",
    "audio/mpeg": "mp3",
    "audio/wav": "wav",
    "audio/x-wav": "wav",
    "audio/ogg": "ogg",
    "audio/webm": "webm",
    "audio/opus": "opus",
    "video/mp4": "mp4",
}

def _dataurl_to_file(data_url: str, orig_name: str | None = None) -> str:
    # data_url: "data:audio/mp4;base64,AAAA..."
    try:
        header, b64 = data_url.split(",", 1)
    except ValueError:
        raise ValueError("data URL 格式錯誤（缺少逗號）。")
    # 取 MIME
    mime = header.split(";")[0].split(":", 1)[-1].strip()
    ext = MIME_EXT.get(mime) or (mimetypes.guess_extension(mime) or "m4a").lstrip(".")
    # 臨時檔名
    fname = orig_name if (orig_name and "." in orig_name) else f"upload_{uuid.uuid4().hex}.{ext}"
    with open(fname, "wb") as f:
        f.write(base64.b64decode(b64))
    return fname

def _extract_effective_path(file_obj) -> str:
    """
    從 Gradio 的 File 輸入（可能是 str / dict / 物件）中，得到真正存在的檔案路徑。
    若沒有實檔，就從 data:URL 產生一個臨時檔。
    """
    # 情況 A：字串（可能是路徑或 data:URL）
    if isinstance(file_obj, str):
        s = file_obj.strip().strip('"')
        if s.startswith("data:"):
            return _dataurl_to_file(s, None)
        if os.path.isfile(s):
            return s
        # 空字串或無效 → 等下嘗試其他來源
    # 情況 B：dict（/gradio_api/call 會送這型）
    if isinstance(file_obj, dict):
        # 優先用 path
        p = str(file_obj.get("path") or "").strip().strip('"')
        if p and os.path.isfile(p):
            return p
        # 不行就用 data
        data = file_obj.get("data")
        if isinstance(data, str) and data.startswith("data:"):
            return _dataurl_to_file(data, file_obj.get("orig_name"))
        # 有些版本可能把真路徑放在 url
        u = str(file_obj.get("url") or "").strip().strip('"')
        if u and os.path.isfile(u):
            return u
    # 情況 C：物件（本機 UI 上傳常見）
    for attr in ("name", "path"):
        p = getattr(file_obj, attr, None)
        if isinstance(p, str):
            s = p.strip().strip('"')
            if os.path.isfile(s):
                return s
    # 物件上有 data:URL？
    data = getattr(file_obj, "data", None)
    if isinstance(data, str) and data.startswith("data:"):
        return _dataurl_to_file(data, getattr(file_obj, "orig_name", None))

    raise FileNotFoundError("無法解析上傳檔案：沒有有效路徑，也沒有 data:URL。")

# ====== 分段處理 ======
def split_audio(path):
    size = os.path.getsize(path)
    if size <= MAX_SIZE:
        return [path]
    audio = AudioSegment.from_file(path)
    n = int(size / MAX_SIZE) + 1
    chunk_ms = len(audio) / n
    parts = []
    for i in range(n):
        fn = f"chunk_{i+1}.wav"
        audio[int(i*chunk_ms):int((i+1)*chunk_ms)].export(fn, format="wav")
        parts.append(fn)
    return parts

# ====== 轉錄核心 ======
def transcribe_core(path, model="whisper-1"):
    # iPhone LINE 常見：mp4（其實是音訊容器）
    if path.lower().endswith(".mp4"):
        fixed = path[:-4] + ".m4a"
        try:
            shutil.copy(path, fixed)
            path = fixed
        except Exception as e:
            print(f"⚠️ mp4→m4a 失敗: {e}")

    chunks = split_audio(path)
    raw = []
    for c in chunks:
        with open(c, "rb") as af:
            txt = client.audio.transcriptions.create(
                model=model,
                file=af,
                response_format="text"
            )
            raw.append(txt)
    raw_txt = "\n".join(raw)

    # 簡轉繁（不意譯）
    conv = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role":"system","content":"你是嚴格的繁體中文轉換器"},
            {"role":"user","content":f"將以下內容轉為台灣繁體，不意譯：\n{raw_txt}"}
        ],
        temperature=0.0
    )
    trad = conv.choices[0].message.content.strip()

    # 摘要（內容多→條列；內容少→一句話）
    summ = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role":"system","content":"你是繁體摘要助手"},
            {"role":"user","content":f"請用台灣繁體中文摘要；內容多則條列重點，內容短則一句話：\n{trad}"}
        ],
        temperature=0.2
    )
    return trad, summ.choices[0].message.content.strip()

# ====== 對外函式（UI / API 共用） ======
def transcribe(password, file):
    if password.strip() != PASSWORD:
        return "❌ 密碼錯誤", "", ""
    if not file:
        return "⚠️ 未選擇檔案", "", ""

    try:
        path = _extract_effective_path(file)
    except Exception as e:
        return f"❌ 檔案解析失敗：{e}", "", ""

    text, summary = transcribe_core(path)
    return "✅ 完成", text, summary

# ====== Gradio UI ======
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("## 🎧 LINE 語音轉錄與摘要（Hugging Face 版）")
    pw = gr.Textbox(label="密碼", type="password")
    f = gr.File(label="上傳音訊檔")
    run = gr.Button("開始轉錄 🚀")
    s = gr.Textbox(label="狀態", interactive=False)
    t = gr.Textbox(label="轉錄結果", lines=10)
    su = gr.Textbox(label="AI 摘要", lines=8)
    # 🔴 關鍵：這個事件關閉 queue → /gradio_api/call/transcribe 直接回結果
    run.click(transcribe, [pw, f], [s, t, su], queue=False)

app = demo

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)