app.py

import os
import tempfile
import requests
import gradio as gr
import torch
import whisper
from gtts import gTTS
from pydub import AudioSegment
from sentence_transformers import SentenceTransformer, util
from transformers import AutoTokenizer, AutoModelForCausalLM
from bs4 import BeautifulSoup
from duckduckgo_search import DDGS
import time

# ------------------------
# 模型初始化
# ------------------------
device = "cuda" if torch.cuda.is_available() else "cpu"
whisper_model = whisper.load_model("base", device=device)
embedder = SentenceTransformer("all-MiniLM-L6-v2")
tokenizer = AutoTokenizer.from_pretrained("gpt2")
lm_model = AutoModelForCausalLM.from_pretrained("gpt2").to(device)

# ------------------------
# 信任網站（含行政院人事行政總處行事曆）
# ------------------------
TRUSTED_SITES = [
    "https://www.dgpa.gov.tw",  # 行政院人事行政總處 行事曆
    "https://www.cna.com.tw",   # 中央通訊社
    "https://www.stust.edu.tw", # 南台科技大學
    "https://www.moi.gov.tw",
    "https://www.taiwan.net.tw",
    "https://publicholidays.tw",
    "https://zh.wikipedia.org",  # 中文維基百科
]

# ------------------------
# 文字轉語音
# ------------------------
def text_to_speech(text):
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as fp:
        tts = gTTS(text=text, lang="zh")
        tts.save(fp.name)
        return fp.name

# ------------------------
# 爬蟲 - 行政院人事行政總處 行事曆
# ------------------------
def crawl_official_calendar(query):
    try:
        url = "https://www.dgpa.gov.tw/holidaycalendar"
        resp = requests.get(url, timeout=10)
        soup = BeautifulSoup(resp.text, "html.parser")
        rows = soup.find_all("tr")
        for row in rows:
            if query in row.text:
                return row.text.strip()
    except Exception as e:
        return None
    return None

# ------------------------
# 網路搜尋 (DuckDuckGo)
# ------------------------
def web_search(query):
    try:
        ddgs = DDGS()
        results = ddgs.text(query, max_results=3)
        answer = ""
        for r in results:
            answer += f"{r['title']} - {r['body']}\n"
        return answer if answer else "查無資料。"
    except Exception:
        return "網路搜尋失敗。"

# ------------------------
# AI 回答
# ------------------------
def ai_answer(question):
    inputs = tokenizer(question, return_tensors="pt").to(device)
    outputs = lm_model.generate(inputs["input_ids"], max_length=150, do_sample=True, top_k=50)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# ------------------------
# 主邏輯
# ------------------------
def qa_system(audio, text_input):
    query = ""
    if text_input:  # 文字優先
        query = text_input.strip()
    elif audio is not None:
        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmpfile:
            audio.export(tmpfile.name, format="wav")
            result = whisper_model.transcribe(tmpfile.name)
            query = result["text"]
    else:
        return "請提供語音或文字輸入。", None, ""

    # 判斷是否包含「南台/南臺」
    if "南台" in query or "南臺" in query:
        # 使用 AI 回答
        start_time = time.time()
        answer = ai_answer(query)
        elapsed = time.time() - start_time
        if elapsed > 120:  # 超過 120 秒 fallback
            web_ans = web_search(query)
            answer += f"\n（補充搜尋結果）{web_ans}"
    else:
        # 節日 or 一般問題，先查可信網站
        start_time = time.time()
        answer = crawl_official_calendar(query)
        if not answer:
            elapsed = time.time() - start_time
            if elapsed > 60:
                answer = web_search(query)
            else:
                answer = "查無相關資料（可信網站）。"

    # 生成語音回覆
    audio_file = text_to_speech(answer)
    return query, answer, audio_file

# ------------------------
# Gradio 介面
# ------------------------
with gr.Blocks() as demo:
    gr.Markdown("# 南臺科技大學語音文字問答系統")
    gr.Markdown("你可以輸入文字或錄音提問，系統會找出答案，並用語音回覆。")

    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(label="文字輸入（可選）", placeholder="請輸入你的問題...")
            audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath", label="請上傳語音檔或錄音")
            submit_btn = gr.Button("Submit")
        with gr.Column():
            text_output = gr.Textbox(label="語音辨識文字")
            answer_output = gr.Textbox(label="AI 回答")
            audio_output = gr.Audio(label="語音回覆", type="filepath")

    submit_btn.click(
        qa_system,
        inputs=[audio_input, text_input],
        outputs=[text_output, answer_output, audio_output]
    )

if __name__ == "__main__":
    demo.launch()