app.py

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

# === 解鎖 Gradio 檔案限制 ===
import gradio.processing_utils as pu
def _dummy_check_allowed(*a, **kw): return True
pu._check_allowed = _dummy_check_allowed
print("🔓 已解除 Gradio 上傳路徑限制")

# === 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 =====")

# === 分段 ===
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
    files = []
    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")
        files.append(fn)
    return files

# === 核心轉錄 ===
def transcribe_core(path, model="whisper-1"):
    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)
    text_list = []
    for f in chunks:
        with open(f, "rb") as af:
            txt = client.audio.transcriptions.create(model=model, file=af, response_format="text")
            text_list.append(txt)
    full_txt = "\n".join(text_list)

    trad = 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{full_txt}"}
        ], temperature=0.0).choices[0].message.content.strip()

    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"用條列或一句話摘要：\n{trad}"}
        ], temperature=0.2).choices[0].message.content.strip()
    return trad, summ

# === 外層驗證 ===
def transcribe(password, file):
    if password.strip() != PASSWORD:
        return "❌ 密碼錯誤", "", ""
    if not file:
        return "⚠️ 未選擇檔案", "", ""

    # 🔒 防呆處理 base64 與錯誤 path
    temp_path = "uploaded_audio.m4a"
    try:
        if hasattr(file, "data") and isinstance(file.data, str) and file.data.startswith("data:audio"):
            base64_str = file.data.split(",")[1]
            with open(temp_path, "wb") as f:
                f.write(base64.b64decode(base64_str))
            file.name = temp_path
        elif os.path.isdir(getattr(file, "name", "")):
            print("⚠️ path 是資料夾，改用 base64")
            base64_str = getattr(file, "data", "").split(",")[1]
            with open(temp_path, "wb") as f:
                f.write(base64.b64decode(base64_str))
            file.name = temp_path
    except Exception as e:
        print(f"⚠️ base64 寫入失敗: {e}")
        return f"❌ 上傳格式錯誤: {e}", "", ""

    text, summary = transcribe_core(file.name)
    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（Base64 安全版）")
    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)
    run.click(transcribe, [pw, f], [s, t, su])

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