Upload 6 files

Dockerfile

@@ -0,0 +1,50 @@
+# 基於支持 CUDA 12.1 的官方 NVIDIA Docker 映像
+FROM nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04
+
+# 設置 DEBIAN_FRONTEND 以跳過交互式的 tzdata 設定
+ENV DEBIAN_FRONTEND=noninteractive
+ENV TZ=Europe/London
+
+# 安裝系統依賴和設置時區
+RUN apt-get update && apt-get install -y \
+    python3 \
+    python3-pip \
+    python3-distutils \
+    ffmpeg \
+    tzdata \
+    && ln -fs /usr/share/zoneinfo/$TZ /etc/localtime \
+    && dpkg-reconfigure --frontend noninteractive tzdata \
+    && rm -rf /var/lib/apt/lists/*
+
+# 創建並切換到新用戶
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+# 設置工作目錄
+WORKDIR /app
+
+# 更新 pip 和安裝构建工具
+RUN pip3 install --no-cache-dir --upgrade pip setuptools wheel
+
+# 安裝 PyTorch 2.1.2 並支援 CUDA 12.1
+RUN pip3 install --no-cache-dir torch==2.1.2+cu121 torchvision==0.16.2+cu121 torchaudio==2.1.2 --extra-index-url https://download.pytorch.org/whl/cu121
+
+# 複製 requirements.txt 並安裝 Python 依賴
+COPY --chown=user requirements.txt .
+RUN pip3 install --no-cache-dir -r requirements.txt
+
+# 下載 spaCy 中文模型
+RUN python3 -m spacy download zh_core_web_md
+
+# 複製應用程式檔案
+COPY --chown=user . /app
+
+# 設置環境變量
+ENV PORT=7860
+
+# 暴露應用運行的端口
+EXPOSE 7860
+
+# 啟動 Flask 應用
+CMD ["python3", "app.py"]

app.py

@@ -0,0 +1,526 @@
+import os
+import json
+import datetime
+import subprocess
+from queue import Queue
+from threading import Thread
+
+import torch
+import yt_dlp
+from faster_whisper import WhisperModel
+from flask import Flask, render_template, request, Response, jsonify
+from openai import OpenAI
+import spacy
+from collections import Counter
+
+import time
+import uuid
+
+import logging
+from logging.handlers import RotatingFileHandler
+from werkzeug.utils import secure_filename
+from collections import deque
+
+# 設置基本日誌配置
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# 創建一個文件處理器，使用 RotatingFileHandler 來限制日誌文件大小
+log_file_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'app.log')
+file_handler = RotatingFileHandler(log_file_path, maxBytes=10*1024*1024, backupCount=5, encoding='utf-8')
+file_handler.setLevel(logging.DEBUG)
+
+# 創建一個控制台處理器
+console_handler = logging.StreamHandler()
+console_handler.setLevel(logging.INFO)
+
+# 創建一個格式器
+formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
+file_handler.setFormatter(formatter)
+console_handler.setFormatter(formatter)
+
+# 將處理器添加到日誌器
+logger.addHandler(file_handler)
+logger.addHandler(console_handler)
+
+# 設置其他模塊的日誌級別
+logging.getLogger("faster_whisper").setLevel(logging.INFO)
+
+os.environ['KMP_DUPLICATE_LIB_OK'] = 'TRUE'
+
+app = Flask(__name__, static_folder='static', static_url_path='/static')
+
+# 讀取設定檔
+current_directory = os.path.dirname(os.path.realpath(__file__))
+config_file_path = os.path.join(current_directory, 'config.json')
+try:
+    with open(config_file_path, 'r', encoding='utf-8') as f:
+        config = json.load(f)
+    logger.info("成功加載配置文件")
+except Exception as e:
+    logger.exception("加載配置文件時發生錯誤")
+    raise
+
+# 設置 OpenAI API 金鑰
+client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+
+# 初始化 SpaCy
+nlp = spacy.load(config['spacy_model'])
+
+# 初始化 Whisper 模型
+model = WhisperModel(config['whisper_model'], device="auto", compute_type=config['whisper_compute_type'])
+
+# 設置 FFmpeg 路徑
+ffmpeg_path = config['ffmpeg_path']
+if ffmpeg_path not in os.environ["PATH"]:
+    os.environ["PATH"] += os.pathsep + ffmpeg_path
+
+def send_sse_message(q, data):
+    q.put_nowait(data)
+
+def clean_filename(filename):
+    return ''.join(c for c in filename if c.isalnum() or c in (' ', '.', '_')).rstrip()
+
+def download_audio(youtube_url, save_directory, q):
+    send_sse_message(q, {"status": "開始下載 YouTube 音頻..."})
+    unique_id = str(uuid.uuid4())[:8]  # 生成一個唯一的識別碼
+    output_filename = f"audio_{unique_id}"
+    output_path = os.path.join(save_directory, output_filename)
+
+    ydl_opts = {
+        'format': 'bestaudio/best',
+        'outtmpl': output_path + ".%(ext)s",
+        'postprocessors': [{
+            'key': 'FFmpegExtractAudio',
+            'preferredcodec': 'mp3',
+            'preferredquality': '192',
+        }],
+        'ffmpeg_location': ffmpeg_path,
+        'quiet': True
+    }
+
+    try:
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            info = ydl.extract_info(youtube_url, download=True)
+            video_title = clean_filename(info.get('title', 'Untitled'))
+
+        # 等待一小段時間，確保文件已經完全寫入
+        time.sleep(2)
+
+        # 檢查生成的文件
+        for file in os.listdir(save_directory):
+            if file.startswith(output_filename) and file.endswith('.mp3'):
+                converted_output_path = os.path.join(save_directory, file)
+                break
+        else:
+            raise FileNotFoundError("無法找到下載的音頻文件")
+
+        send_sse_message(q, {"status": f"音頻下載完成: {video_title}"})
+        return converted_output_path, video_title
+    except Exception as e:
+        send_sse_message(q, {"status": f"下載音頻時發生錯誤: {str(e)}"})
+        raise
+
+def process_local_video(video_path, save_directory, q):
+    send_sse_message(q, {"status": "正在處理本地視頻..."})
+    video_title = os.path.splitext(os.path.basename(video_path))[0]
+    output_path = os.path.join(save_directory, f"{video_title}_audio.mp3")
+
+    ffmpeg_command = [
+        os.path.join(ffmpeg_path, 'ffmpeg'),  # 使用完整路徑
+        '-i', video_path,
+        '-vn',  # 禁用視頻
+        '-acodec', 'libmp3lame',  # 使用 MP3 編碼器
+        '-q:a', '2',  # 音頻質量，2 是很好的質量
+        output_path
+    ]
+
+    logger.info(f"FFmpeg 命令: {' '.join(ffmpeg_command)}")
+    logger.info(f"輸入視頻路徑: {video_path}")
+    logger.info(f"輸出音頻路徑: {output_path}")
+
+    try:
+        # 檢查輸入文件是否存在
+        if not os.path.exists(video_path):
+            raise FileNotFoundError(f"輸入視頻文件不存在: {video_path}")
+
+        # 檢查輸出目錄是否可寫
+        if not os.access(os.path.dirname(output_path), os.W_OK):
+            raise PermissionError(f"沒有寫入權限: {os.path.dirname(output_path)}")
+
+        result = subprocess.run(ffmpeg_command, check=True, capture_output=True, text=True)
+        logger.info(f"FFmpeg 輸出: {result.stdout}")
+        send_sse_message(q, {"status": f"本地視頻處理完成: {video_title}"})
+        return output_path, video_title
+    except subprocess.CalledProcessError as e:
+        error_message = f"處理本地視頻時出錯: {e}\n\nFFmpeg 輸出:\n{e.stdout}\n\nFFmpeg 錯誤:\n{e.stderr}"
+        logger.error(error_message)
+        send_sse_message(q, {"status": "錯誤", "error": error_message})
+        raise
+    except Exception as e:
+        error_message = f"處理本地視頻時出現意外錯誤: {str(e)}"
+        logger.error(error_message)
+        send_sse_message(q, {"status": "錯誤", "error": error_message})
+        raise
+
+def generate_transcript(audio_path, video_title, q):
+    send_sse_message(q, {"status": "開始音頻轉錄..."})
+    segments, info = model.transcribe(
+        audio_path, 
+        beam_size=config['whisper_beam_size'],
+        language=config['whisper_language'],
+        temperature=config['whisper_temperature'],
+        initial_prompt=video_title,
+        repetition_penalty=2,
+        condition_on_previous_text=False
+    )
+    transcript = "\n".join([segment.text for segment in segments])
+    send_sse_message(q, {"status": f"音頻轉錄完成，檢測到的語言: {info.language}", "transcript": transcript})
+    return transcript
+
+def smart_split_transcript(transcript, q):
+    send_sse_message(q, {"status": "開始智能分割轉錄文本..."})
+    doc = nlp(transcript)
+    segments = []
+    current_segment = ""
+    max_length = 1024
+
+    for sent in doc.sents:
+        if len(current_segment) + len(sent.text) <= max_length:
+            current_segment += " " + sent.text
+        else:
+            if current_segment:
+                segments.append(current_segment.strip())
+            current_segment = sent.text
+
+    if current_segment:
+        segments.append(current_segment.strip())
+
+    send_sse_message(q, {"status": f"轉錄文本分割完成，共 {len(segments)} 個段落"})
+    return segments
+
+def extract_keywords_and_entities(text):
+    doc = nlp(text)
+    keywords = [token.lemma_ for token in doc if not token.is_stop and not token.is_punct]
+    keyword_freq = Counter(keywords).most_common(5)
+    entities = [(ent.text, ent.label_) for ent in doc.ents]
+    return [keyword for keyword, _ in keyword_freq], entities
+
+def process_youtube_description(description):
+    prompt = f"""請處理以下 YouTube 影片描述，移除所有渠道宣傳內容後，保留原文。
+
+描述內容：
+{description}"""
+
+    response = client.chat.completions.create(
+        model=config['openai_model'],
+        messages=[{"role": "system", "content": prompt}],
+        temperature=0.1,
+        max_tokens=500
+    )
+    
+    processed_description = response.choices[0].message.content.strip()
+    
+    # 在終端機打印處理後的描述
+    print("處理後的 YouTube 描述：")
+    print(processed_description)
+    print("------------------------")
+    
+    return processed_description
+
+def get_openai_summary(segment, video_title, is_final_summary, keywords, entities, processed_description, q):
+    if is_final_summary:
+        prompt = f"""以下是YouTube視頻'{video_title}'的多個段落摘要。請生成一個深入且全面的最終摘要，盡力保留主要內容、資訊細節、關鍵點和結論。摘要應該是連貫的、有條理的、詳細的，並且避免重複信息。在內容結尾，加入能夠方便搜尋器和 SEO 找到的 3 個 Hash Tag。請用繁體中文(香港)回應。
+
+影片描述提供的可靠資訊 (請特別使用來補充和糾正摘要中的信息，尤其是戈人名或專有名詞)：
+{processed_description}
+
+
+
+以下是待處理的摘要內容：
+{segment}"""
+    else:
+        keywords_str = ", ".join(keywords)
+        entities_str = ", ".join([f"{text}({label})" for text, label in entities])
+        prompt = f"""以下內容是YouTube視頻的部份字幕文本，每行以短句顯示，閱讀時需要將多行組合一起才是一句完整的句子，偶爾會出現音譯的錯別字，請修正。內容主題是關於：'{video_title}'，其中包含的關鍵詞有：{keywords_str}，和以下的NER實體：{entities_str}。
+
+影片描述提供的可靠資訊 (請特別使用來補充和糾正摘要中的信息，尤其是戈人名或專有名詞)：
+{processed_description}
+
+請根據每個NER實體的意思，以及上述描述資訊，以不少於 200 字的繁體中文(香港) 重組文章段落。目標是盡量抽取��主題有關的所有觀點、事件、案例、學問、步驟、方法、時間、人物、數據、名詞的基礎資料，建構成一篇連貫的、全面的、詳細的紀錄。請特別注意使用描述資訊來糾正可能的錯誤，尤其是人名和地名。忽略重複的、單純抒發個人情緒的訊息、與 Youtuber 個人宣傳的訊息。
+
+你要處理的內容如下：
+{segment}"""
+    
+    response = client.chat.completions.create(
+        model=config['openai_model'],
+        messages=[{"role": "system", "content": prompt}],
+        temperature=0.6,
+        max_tokens=1000
+    )
+    
+    summary = response.choices[0].message.content.strip()
+    return summary
+
+def save_summary(text, video_title, url_or_path, save_directory):
+    current_time = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+    cleaned_title = clean_filename(video_title)[:20]
+    summary_file_name = f"GPT_Summary_{cleaned_title}_{current_time}.txt"
+    summary_file_path = os.path.join(save_directory, summary_file_name)
+
+    # 移除文本開頭可能存在的影片名稱和 URL/路徑信息
+    lines = text.split('\n')
+    if lines[0].startswith("影片名稱：") and lines[1].startswith("網址或路徑："):
+        text = '\n'.join(lines[2:])
+
+    summary_text = f"影片名稱：\"{video_title}\"\n網址或路徑：\"{url_or_path}\"\n\n{text}"
+
+    with open(summary_file_path, "w", encoding="utf-8") as file:
+        file.write(summary_text)
+
+def save_transcript(transcript, video_title, url_or_path, save_directory):
+    current_time = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+    cleaned_title = clean_filename(video_title)[:20]
+    transcript_file_name = f"Transcript_{cleaned_title}_{current_time}.txt"
+    transcript_file_path = os.path.join(save_directory, transcript_file_name)
+
+    with open(transcript_file_path, "w", encoding="utf-8") as file:
+        file.write(f"影片名稱：\"{video_title}\"\n網址或路徑：\"{url_or_path}\"\n\n{transcript}")
+
+    logger.info(f"轉錄文本已保存至 {transcript_file_path}")
+
+def save_segment_summary(summary_text, segment_index, video_title, save_directory):
+    current_time = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+    cleaned_title = clean_filename(video_title)[:20]
+    summary_file_name = f"Segment_Summary_{cleaned_title}_{segment_index}_{current_time}.txt"
+    summary_file_path = os.path.join(save_directory, summary_file_name)
+    
+    with open(summary_file_path, "w", encoding="utf-8") as file:
+        file.write(summary_text)
+    
+    logger.info(f"段落摘要已保存至 {summary_file_path}")
+
+def process_video(url_or_path, q, local_video_description=''):
+    try:
+        logger.info(f"開始處理視頻: {url_or_path}")
+        save_directory = config['save_directory']
+
+        processed_description = ""
+        if url_or_path.startswith('http'):
+            # YouTube URL 處理邏輯保持不變
+            logger.info("檢測到 YouTube URL，開始獲取視頻信息")
+            ydl_opts = {'quiet': True}
+            with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+                video_info = ydl.extract_info(url_or_path, download=False)
+            
+            video_data = {
+                'title': video_info['title'],
+                'duration': str(datetime.timedelta(seconds=video_info['duration'])),
+                'view_count': video_info['view_count'],
+                'like_count': video_info.get('like_count', 'N/A'),
+                'description': video_info['description']
+            }
+            send_sse_message(q, {"status": "獲取到視頻信息", "video_info": video_data})
+
+            # 處理 YouTube 描述
+            raw_description = video_info['description']
+            processed_description = process_youtube_description(raw_description)
+
+            logger.info("開始下載 YouTube 音頻")
+            audio_path, video_title = download_audio(url_or_path, save_directory, q)
+        else:
+            logger.info("檢測到本地文件路徑，開始處理本地視頻")
+            audio_path, video_title = process_local_video(url_or_path, save_directory, q)
+            processed_description = local_video_description if local_video_description else "這是一個本地視頻文件，用戶沒有提供視頻描述。"
+
+        if not audio_path or not os.path.exists(audio_path):
+            raise FileNotFoundError(f"音頻文件不存在: {audio_path}")
+
+        logger.info("開始生成轉錄文本")
+        transcript = generate_transcript(audio_path, video_title, q)
+        
+        # 保存轉錄文本
+        save_transcript(transcript, video_title, url_or_path, save_directory)
+        
+        logger.info("開始分割轉錄文本")
+        segments = smart_split_transcript(transcript, q)
+
+        all_summaries = []
+        for i, segment in enumerate(segments, start=1):
+            logger.info(f"開始為文本段 {i}/{len(segments)} 生成���要")
+            send_sse_message(q, {"status": f"正在為文本段 {i}/{len(segments)} 生成摘要..."})
+            keywords, entities = extract_keywords_and_entities(segment)
+            segment_summary = get_openai_summary(segment, video_title, False, keywords, entities, processed_description, q)
+            if segment_summary:
+                all_summaries.append(segment_summary)
+                save_segment_summary(segment_summary, i, video_title, save_directory)
+                send_sse_message(q, {"status": f"段落 {i} 摘要完成", "summary": segment_summary})
+
+        logger.info("開始生成最終摘要")
+        send_sse_message(q, {"status": "正在生成最終摘要..."})
+        all_summaries_text = "\n\n".join(all_summaries)
+        final_summary = get_openai_summary(all_summaries_text, video_title, True, [], [], processed_description, q)
+        # 將最終摘要添加到 summary_versions
+        summary_versions.append(final_summary)
+        # 修改這裡：發送包含版本信息的最終摘要
+        send_sse_message(q, {
+            "status": "處理完成", 
+            "final_summary": final_summary,
+            "version": 0,
+            "total_versions": len(summary_versions)
+        })
+        
+        # 添加影片名稱和 URL/路徑到最終摘要
+        final_summary_with_info = f'影片名稱："{video_title}"\n網址或路徑："{url_or_path}"\n\n{final_summary}'
+        
+        send_sse_message(q, {"status": "處理完成", "final_summary": final_summary_with_info})
+
+        # 保存最終摘要
+        logger.info("保存最終摘要")
+        save_summary(final_summary_with_info, video_title, url_or_path, save_directory)
+
+        # 刪除臨時音頻文件
+        if os.path.exists(audio_path):
+            try:
+                os.remove(audio_path)
+                logger.info("臨時音頻文件已刪除")
+                send_sse_message(q, {"status": "臨時音頻文件已刪除"})
+            except Exception as e:
+                logger.error(f"無法刪除臨時音頻文件: {str(e)}")
+                send_sse_message(q, {"status": f"無法刪除臨時音頻文件: {str(e)}"})
+
+        # 如果是本地上傳的 .mp4 文件，刪除臨時文件
+        if not url_or_path.startswith('http') and url_or_path.lower().endswith('.mp4'):
+            try:
+                os.remove(url_or_path)
+                logger.info("臨時上傳的 .mp4 文件已刪除")
+                send_sse_message(q, {"status": "臨時上傳的 .mp4 文件已刪除"})
+            except Exception as e:
+                logger.error(f"無法刪除臨時上傳的 .mp4 文件: {str(e)}")
+                send_sse_message(q, {"status": f"無法刪除臨時上傳的 .mp4 文件: {str(e)}"})
+
+        logger.info("視頻處理完成")
+
+    except Exception as e:
+        logger.exception("處理視頻時發生錯誤")
+        send_sse_message(q, {"status": f"錯誤: {str(e)}"})
+
+
+# 在全局變量部分添加：
+refinement_count = 0
+max_refinement_count = config.get('max_refinement_count', 5)  # 使用 get 方法，如果 config.json 配置中沒有，則使用默認值 5
+summary_versions = deque(maxlen=max_refinement_count + 1)
+
+# 添加新的函數：
+def refine_final_summary(original_summary, user_feedback, video_title, processed_description):
+    prompt = f"""你是一個專業的廣東話視頻內容摘要編輯。請根據用戶的反饋，改進以下內容摘要。標題是"{video_title}"。
+
+原始摘要：
+{original_summary}
+
+用戶反饋：
+{user_feedback}
+
+
+請遵循以下指引：
+1. 仔細閱讀原始摘要和用戶反饋，以用戶反饋的指示作為優先原則。
+2. 根據用戶反饋，補充、修正在原始摘要內，任何錯誤或不準確的資訊，確保摘要全面涵蓋主題內容。
+3. 保留原始摘要中準確和重要的部分。
+4. 確保摘要邏輯清晰，結構完整，易於閱讀理解。
+5. 如有必要，重新組織摘要結構以提高清晰度和連貫性。
+6. 保留原有的 Hash Tag（如果有的話），或根據更新後的內容調整 Hash Tag。
+
+請生成最終摘要，確保其準確、全面、連貫，並符合用戶的反饋意見。"""
+
+    response = client.chat.completions.create(
+        model=config['openai_model'],
+        messages=[{"role": "system", "content": prompt}],
+        temperature=0.8,
+        max_tokens=1000
+    )
+    
+    refined_summary = response.choices[0].message.content.strip()
+    return refined_summary
+
+# 添加新的路由：
+@app.route('/refine_summary', methods=['POST'])
+def refine_summary():
+    global refinement_count
+    data = request.json
+    #logger.info(f"Received refinement request: {data}") #{'original_summary': .... 'user_feedback': .... 'video_title':...'video_url'...'processed_description'...
+    original_summary = data['original_summary']
+    user_feedback = data['user_feedback']
+    video_title = data['video_title']
+    video_url = data['video_url']
+    processed_description = data['processed_description']
+
+    if refinement_count >= config['max_refinement_count']:
+        return jsonify({"error": "已達到最大重新生成次數"}), 400
+
+    refined_summary = refine_final_summary(original_summary, user_feedback, video_title, processed_description)
+    refinement_count += 1
+    
+    # 添加視頻信息到摘要
+    refined_summary_with_info = f"影片名稱：{video_title}\n網址或路徑：{video_url}\n\n{refined_summary}"
+    
+    logger.info(f"Sending refined summary: {refined_summary_with_info}")
+    return jsonify({
+        "refined_summary": refined_summary_with_info,
+        "version": refinement_count,
+        "total_versions": refinement_count + 1
+    })
+
+
+
+@app.route('/')
+def index():
+    return render_template('index.html')
+
+@app.route('/process', methods=['POST'])
+def process():
+    try:
+        url_or_path = request.form.get('url_or_path')
+        
+        if not url_or_path:
+            return jsonify({"error": "No URL or path provided"}), 400
+        
+        if url_or_path.startswith('http'):
+            # YouTube URL 處理邏輯保持不變
+            pass
+        else:
+            # 本地文件處理
+            if 'file' not in request.files:
+                return jsonify({"error": "No file uploaded"}), 400
+            file = request.files['file']
+            if file.filename == '':
+                return jsonify({"error": "No file selected"}), 400
+            if file:
+                filename = secure_filename(file.filename)
+                file_path = os.path.join(config['save_directory'], filename)
+                file.save(file_path)
+                url_or_path = file_path
+        
+        # 獲取本地視頻描述
+        local_video_description = request.form.get('localVideoDescription', '')
+        
+        logger.info(f"處理文件: {url_or_path}")
+        
+        q = Queue()
+        thread = Thread(target=process_video, args=(url_or_path, q, local_video_description))
+        thread.start()
+        return Response(event_stream(q), content_type='text/event-stream')
+    except Exception as e:
+        error_message = f"處理請求時出現錯誤: {str(e)}"
+        logger.error(error_message)
+        return jsonify({"error": error_message}), 500
+
+def event_stream(q):
+    while True:
+        message = q.get()
+        yield f"data: {json.dumps(message)}\n\n"
+        if message.get('status') == '處理完成' or message.get('status').startswith('錯誤'):
+            break
+
+if __name__ == '__main__':
+    port = int(os.environ.get('PORT', 5000))
+    app.run(host='0.0.0.0', port=port)

config.json

@@ -0,0 +1,14 @@
+{
+  "save_directory": "D:\\_Adam_Projects\\AI\\myWorkspace_python\\YouTube_Caption_Generator\\HF_space\\output",
+  "ffmpeg_path": "D:\\_Adam_Projects\\_Toolbox\\ffmpeg\\bin",
+  "spacy_model": "zh_core_web_md",
+  "whisper_model": "large-v3",
+  "whisper_compute_type": "int8_float16",
+  "whisper_language": "zh",
+  "whisper_beam_size": 8,
+  "whisper_temperature" : 0.4,
+  "audio_start_percent":0,
+  "audio_end_percent":100,
+  "openai_model": "gpt-4o-mini",
+  "max_refinement_count": 5
+}

requirements.txt

@@ -0,0 +1,9 @@
+Flask==3.0.2
+yt-dlp==2024.10.7
+faster-whisper==1.0.3
+openai==1.51.2
+spacy==3.7.5
+numpy==1.26.4
+pydantic==2.9.2
+typer==0.12.1
+Werkzeug==3.0.1

static/main.js

@@ -0,0 +1,360 @@
+let processingMethod = '';
+let currentSummaryVersion = 0;
+let totalSummaryVersions = 1;
+let summaryVersions = [];
+let currentVideoTitle = '';
+let currentVideoUrl = '';
+
+window.updateSummaryDisplay = function() {
+    if (summaryVersions.length > 0 && summaryVersions[currentSummaryVersion]) {
+        document.getElementById('finalSummary').innerHTML = linkifyText(summaryVersions[currentSummaryVersion]);
+        document.getElementById('summaryVersion').textContent = `版本：${currentSummaryVersion + 1}/${totalSummaryVersions}`;
+        document.getElementById('prevVersion').disabled = currentSummaryVersion === 0;
+        document.getElementById('nextVersion').disabled = currentSummaryVersion === totalSummaryVersions - 1;
+    } else {
+        console.error('No summary available for the current version');
+        document.getElementById('finalSummary').innerHTML = '暫無可用摘要';
+    }
+}
+document.addEventListener('DOMContentLoaded', (event) => {
+
+    document.getElementById('refineSummary').addEventListener('click', showSummaryLightbox);
+    document.getElementById('submitFeedback').addEventListener('click', submitFeedback);
+    document.getElementById('prevVersion').addEventListener('click', () => changeSummaryVersion(-1));
+    document.getElementById('nextVersion').addEventListener('click', () => changeSummaryVersion(1));
+    
+    // 點擊 Lightbox 外部關閉
+    document.getElementById('summaryLightbox').addEventListener('click', function(e) {
+        if (e.target === this) {
+            hideSummaryLightbox();
+        }
+    });
+
+    function updateSummaryDisplay() {
+        console.log("Updating summary display", { currentSummaryVersion, totalSummaryVersions, summaryVersions });
+        if (summaryVersions.length > 0 && summaryVersions[currentSummaryVersion]) {
+            document.getElementById('finalSummary').innerHTML = linkifyText(summaryVersions[currentSummaryVersion]);
+            document.getElementById('summaryVersion').textContent = `版本：${currentSummaryVersion + 1}/${totalSummaryVersions}`;
+            document.getElementById('prevVersion').disabled = currentSummaryVersion === 0;
+            document.getElementById('nextVersion').disabled = currentSummaryVersion === totalSummaryVersions - 1;
+        } else {
+            console.error('No summary available for the current version');
+            document.getElementById('finalSummary').innerHTML = '暫無可用摘要';
+        }
+    }
+    
+    function showSummaryLightbox() {
+        document.getElementById('summaryLightbox').classList.remove('hidden');
+        document.getElementById('submitFeedback').disabled = false;
+        document.getElementById('feedbackLoading').classList.add('hidden');
+    }
+
+    function hideSummaryLightbox() {
+        document.getElementById('summaryLightbox').classList.add('hidden');
+    }
+
+    function submitFeedback() {
+    console.log("開始提交反饋");
+    console.log("當前摘要版本:", currentSummaryVersion);
+    console.log("總摘要版本數:", totalSummaryVersions);
+    const feedback = document.getElementById('userFeedback').value;
+    const originalSummary = summaryVersions[currentSummaryVersion];
+    const videoInfoContent = document.getElementById('videoInfoContent').textContent;
+    let videoTitle = currentVideoTitle;
+    let videoUrl = currentVideoUrl;
+
+    // 顯示 loading 效果，禁用提交按鈕
+    document.getElementById('submitFeedback').disabled = true;
+    document.getElementById('feedbackLoading').classList.remove('hidden');
+    
+    // 禁用點擊外部關閉
+    document.getElementById('summaryLightbox').style.pointerEvents = 'none';
+
+    // 移除原始摘要中的視頻信息
+    const summaryWithoutInfo = originalSummary.split('\n').slice(2).join('\n').trim();
+
+    console.log("Submitting feedback:", {
+        original_summary: summaryWithoutInfo,
+        user_feedback: feedback,
+        video_title: videoTitle,
+        video_url: videoUrl
+    });
+
+    fetch('/refine_summary', {
+        method: 'POST',
+        headers: {
+            'Content-Type': 'application/json',
+        },
+        body: JSON.stringify({
+            original_summary: summaryWithoutInfo,
+            user_feedback: feedback,
+            video_title: videoTitle,
+            video_url: videoUrl,
+            processed_description: document.getElementById('videoInfoContent').textContent
+        })
+    })
+    .then(response => {
+        if (!response.ok) {
+            throw new Error(`HTTP error! status: ${response.status}`);
+        }
+        return response.json();
+    })
+    .then(data => {
+        console.log("Received data from server:", data);
+        if (data.error) {
+            throw new Error(data.error);
+        }
+        if (data.refined_summary) {
+            console.log("收到新的摘要:", data.refined_summary);
+            
+            // 添加新摘要到陣列
+            summaryVersions.push(data.refined_summary);
+            
+            // 更新當前版本為最新版本
+            currentSummaryVersion = summaryVersions.length - 1;
+            
+            // 更新總版本數
+            totalSummaryVersions = summaryVersions.length;
+            
+            console.log("更新後的版本信息:", {
+                currentSummaryVersion: currentSummaryVersion,
+                totalSummaryVersions: totalSummaryVersions,
+                summaryVersionsLength: summaryVersions.length
+            });
+        
+            // 更新顯示
+            window.updateSummaryDisplay();
+        
+            console.log("摘要顯示已更新");
+        
+            // 隱藏反饋輸入框
+            hideSummaryLightbox();
+        } else {
+            console.error("收到的摘要為空");
+            throw new Error("收到的摘要為空");
+        }
+        hideSummaryLightbox();
+    })
+    .catch(error => {
+        console.error('Error during feedback submission:', error);
+        alert('提交反饋時出錯: ' + error.message);
+    })
+    .finally(() => {
+        // 恢復提交按鈕，隱藏 loading 效果
+        document.getElementById('submitFeedback').disabled = false;
+        document.getElementById('feedbackLoading').classList.add('hidden');
+        // 恢復點擊外部關閉
+        document.getElementById('summaryLightbox').style.pointerEvents = 'auto';
+    });
+}
+    
+    
+
+    function changeSummaryVersion(delta) {
+        currentSummaryVersion += delta;
+        if (currentSummaryVersion < 0) currentSummaryVersion = 0;
+        if (currentSummaryVersion >= totalSummaryVersions) currentSummaryVersion = totalSummaryVersions - 1;
+        window.updateSummaryDisplay(); // 使用 window.updateSummaryDisplay
+    }
+    
+
+    document.getElementById('youtubeBtn').addEventListener('click', () => {
+        processingMethod = 'youtube';
+        document.getElementById('youtubeInput').classList.remove('hidden');
+        document.getElementById('localVideoInput').classList.add('hidden');
+    });
+
+    document.getElementById('localVideoBtn').addEventListener('click', () => {
+        processingMethod = 'local';
+        document.getElementById('localVideoInput').classList.remove('hidden');
+        document.getElementById('youtubeInput').classList.add('hidden');
+    });
+
+    document.getElementById('startProcessing').addEventListener('click', () => {
+        let url_or_path = '';
+        if (processingMethod === 'youtube') {
+            url_or_path = document.getElementById('youtubeUrl').value;
+        } else if (processingMethod === 'local') {
+            const fileInput = document.getElementById('localVideoFile');
+            if (fileInput.files.length > 0) {
+                url_or_path = fileInput.files[0].name;
+            }
+        }
+
+        if (!url_or_path) {
+            alert('請選擇處理方式並輸入 URL 或選擇文件');
+            return;
+        }
+
+        if (hasExistingContent()) {
+            if (confirm('開始新的處理任務將清空當前內容。是否繼續？')) {
+                clearPreviousContent();
+                startProcessing(url_or_path);
+            }
+        } else {
+            startProcessing(url_or_path);
+        }
+    });
+
+    document.getElementById('toggleTranscript').addEventListener('click', () => {
+        const transcript = document.getElementById('transcript');
+        transcript.style.display = transcript.style.display === 'none' ? 'block' : 'none';
+    });
+});
+
+function hasExistingContent() {
+    return !document.getElementById('videoInfo').classList.contains('hidden') ||
+        !document.getElementById('summary').classList.contains('hidden') ||
+        !document.getElementById('segmentSummaries').classList.contains('hidden') ||
+        document.getElementById('transcript').textContent.trim() !== '';
+}
+
+function clearPreviousContent() {
+    // 清空 UI 內容
+    document.getElementById('videoInfo').classList.add('hidden');
+    document.getElementById('videoInfoContent').textContent = '';
+    document.getElementById('status').classList.add('hidden');
+    document.getElementById('status').textContent = '';
+    document.getElementById('summary').classList.add('hidden');
+    document.getElementById('finalSummary').innerHTML = '';
+    document.getElementById('segmentSummaries').classList.add('hidden');
+    document.getElementById('summaries').innerHTML = '';
+    document.getElementById('transcript').textContent = '';
+
+    // 重置全局變量
+    currentSummaryVersion = 0;
+    totalSummaryVersions = 1;  // 重置為 1
+    summaryVersions = [];
+    
+    // 重置版本顯示
+    document.getElementById('summaryVersion').textContent = '版本：1/1';
+    
+    // 禁用版本切換按鈕
+    document.getElementById('prevVersion').disabled = true;
+    document.getElementById('nextVersion').disabled = true;
+    
+    // 清空反饋輸入框
+    document.getElementById('userFeedback').value = '';
+}
+
+function startProcessing(url_or_path) {
+    clearPreviousContent(); 
+    currentVideoTitle = '';
+    currentVideoUrl = url_or_path;
+    document.getElementById('status').classList.remove('hidden');
+    document.getElementById('status').textContent = '⌛️ 處理中...';
+    
+    const formData = new FormData();
+    formData.append('url_or_path', url_or_path);
+    if (processingMethod === 'local') {
+        const fileInput = document.getElementById('localVideoFile');
+        if (fileInput.files.length > 0) {
+            formData.append('file', fileInput.files[0]);
+            currentVideoTitle = fileInput.files[0].name;
+        }
+        formData.append('localVideoDescription', document.getElementById('localVideoDescription').value);
+    } else {
+        currentVideoTitle = ''; // 將在收到視頻信息時更新
+    }
+
+    fetch('/process', {
+        method: 'POST',
+        body: formData
+    })
+    .then(response => {
+        const reader = response.body.getReader();
+        const decoder = new TextDecoder();
+        
+        function readStream() {
+            return reader.read().then(({ done, value }) => {
+                if (done) {
+                    console.log('Stream complete');
+                    return;
+                }
+                const chunk = decoder.decode(value);
+                handleServerSentEvent(chunk);
+                return readStream();
+            });
+        }
+        
+        return readStream();
+    })
+    .catch(error => {
+        console.error('Error:', error);
+        document.getElementById('status').textContent = `錯誤: ${error.message}`;
+    });
+}
+
+function handleServerSentEvent(chunk) {
+    const lines = chunk.split('\n');
+    lines.forEach(line => {
+        if (line.startsWith('data: ')) {
+            const data = JSON.parse(line.slice(6));
+            updateUI(data);
+        }
+    });
+}
+
+function updateUI(data) {
+    document.getElementById('status').textContent = data.status;
+    
+    
+    if (data.video_info) {
+        document.getElementById('videoInfo').classList.remove('hidden');
+        let videoInfoContent = `標題：${data.video_info.title}\n`;
+        videoInfoContent += `時長：${data.video_info.duration}\n`;
+        videoInfoContent += `觀看次數：${data.video_info.view_count}\n`;
+        videoInfoContent += `喜歡數：${data.video_info.like_count}\n`;
+        videoInfoContent += `描述：${data.video_info.description}`;
+        
+        if (videoInfoContent.length > 300) {
+            videoInfoContent = videoInfoContent.substring(0, 300) + '... ...';
+        }
+        
+        document.getElementById('videoInfoContent').textContent = videoInfoContent;
+        // 更新當前視頻標題
+        currentVideoTitle = data.video_info.title;
+    }
+    
+    console.log("Received data:", data);
+
+    if (data.final_summary) {
+        console.log("Updating final summary");
+        document.getElementById('summary').classList.remove('hidden');
+        const summaryWithInfo = `影片名稱：${currentVideoTitle}\n網址或路徑：${currentVideoUrl}\n\n${data.final_summary}`;
+
+        /* // 從視頻信息中獲取標題和 URL/路徑
+        const videoTitle = document.getElementById('videoInfoContent').textContent.split('\n')[0].replace('標題：', '').trim();
+        const videoUrl = document.getElementById('youtubeUrl').value || document.getElementById('localVideoFile').value;
+        
+        // 在摘要前添加影片名稱和 URL/路徑信息
+        const summaryWithInfo = `影片名稱：${videoTitle}\n網址或路徑：${videoUrl}\n\n${data.final_summary}`; */
+        
+        summaryVersions = [summaryWithInfo];
+        currentSummaryVersion = 0;
+        totalSummaryVersions = 1;  // 重置為 1
+        console.log("Calling updateSummaryDisplay");
+        window.updateSummaryDisplay();
+        console.log("updateSummaryDisplay called");
+    }
+    
+    if (data.summary) {
+        document.getElementById('segmentSummaries').classList.remove('hidden');
+        document.getElementById('summaries').innerHTML += linkifyText(data.summary) + '\n\n';
+    }
+    
+    if (data.transcript) {
+        document.getElementById('transcript').textContent = data.transcript;
+    }
+    
+    if (data.error) {
+        alert(data.error);
+    }
+}
+
+function linkifyText(text) {
+    const urlRegex = /(https?:\/\/[^\s]+)/g;
+    return text.replace(urlRegex, function(url) {
+        return `<a href="${url}" target="_blank" class="text-blue-600 hover:underline">${url}</a>`;
+    });
+}

templates/index.html

@@ -0,0 +1,104 @@
+<!DOCTYPE html>
+<html lang="zh-HK">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>廣東話 Youtube 內容總結工具</title>
+    <script src="https://cdn.tailwindcss.com"></script>
+    <style>
+        .summary-content {
+            white-space: pre-wrap;
+            word-wrap: break-word;
+        }
+        .transcript-content {
+            display: none;
+            white-space: pre-wrap;
+            word-wrap: break-word;
+        }
+        .btn-custom {
+            transition: background-color 0.3s;
+        }
+        .btn-custom:hover {
+            opacity: 0.9;
+        }
+    </style>
+</head>
+<body class="bg-gray-100">
+    <div class="container mx-auto px-4 py-8">
+        <h1 class="text-3xl font-bold mb-4">廣東話 Youtube 內容總結工具</h1>
+        <p class="text-sm text-gray-500 mb-4">by Adam Chan. 2024-10 (Beta-0.8) 😃如果這工具對你有用，歡迎發個訊息👍給作者以示鼓勵。</p>
+        <p class="mb-6">這工具能自動下載 YouTube 視頻音頻，轉錄為文字，並生成詳細摘要。它支持長視頻，使用先進的 AI 模型進行轉錄和摘要生成。無論是研究、學習還是內容創作，這工具都能幫您快速掌握視頻核心內容，節省寶貴時間。現在還能利用視頻描述來提高摘要的準確性！</p>
+
+        <div class="mb-6">
+            <h2 class="text-xl font-semibold mb-2">選擇處理方式：</h2>
+            <div class="flex space-x-4">
+                <button id="youtubeBtn" class="btn-custom px-4 py-2 bg-[#E57373] text-white rounded">YouTube 視頻</button>
+                <button id="localVideoBtn" class="btn-custom px-4 py-2 bg-[#64B5F6] text-white rounded">本地視頻文件</button>
+            </div>
+        </div>
+
+        <div id="youtubeInput" class="mb-6 hidden">
+            <label for="youtubeUrl" class="block mb-2">輸入 YouTube URL：</label>
+            <input type="text" id="youtubeUrl" class="w-full p-2 border rounded" placeholder="https://www.youtube.com/watch?v=...">
+        </div>
+
+        <div id="localVideoInput" class="mb-6 hidden">
+            <label for="localVideoFile" class="block mb-2">選擇本地視頻文件（僅支持 .mp4）：</label>
+            <input type="file" id="localVideoFile" accept=".mp4" class="w-full p-2 border rounded">
+            <textarea id="localVideoDescription" class="w-full p-2 border rounded mt-2" rows="4" placeholder="請輸入視頻描述（可選）"></textarea>
+        </div>
+
+        <button id="startProcessing" class="btn-custom px-6 py-3 bg-[#81C784] text-white rounded font-bold">開始處理</button>
+
+        <div id="videoInfo" class="mt-6 hidden">
+            <h3 class="text-lg font-semibold mb-2">📺 視頻信息：</h3>
+            <div id="videoInfoContent" class="summary-content p-4 bg-white rounded shadow"></div>
+        </div>
+
+        <div id="status" class="mt-6 p-4 bg-blue-100 rounded hidden"></div>
+
+        <div id="summary" class="mt-6 hidden">
+            <h3 class="text-lg font-semibold mb-2">📍 最終摘要：</h3>
+            <div id="finalSummary" class="summary-content p-4 bg-white rounded shadow"></div>
+            <div class="mt-4 flex justify-between items-center">
+                <button id="refineSummary" class="bg-blue-500 hover:bg-blue-700 text-white font-bold py-2 px-4 rounded">
+                    重新生成摘要
+                </button>
+                <button id="prevVersion" class="bg-gray-300 hover:bg-gray-400 text-gray-800 font-bold py-2 px-4 rounded">
+                    上一個版本
+                </button>
+                <button id="nextVersion" class="bg-gray-300 hover:bg-gray-400 text-gray-800 font-bold py-2 px-4 rounded">
+                    下一個版本
+                </button>
+                <span id="summaryVersion">版本：1/1</span>
+            </div>
+        </div>
+
+        <!-- 添加 Lightbox -->
+        <div id="summaryLightbox" class="fixed inset-0 bg-gray-600 bg-opacity-50 overflow-y-auto h-full w-full hidden">
+            <div class="relative top-20 mx-auto p-5 border w-2/3 max-w-2xl shadow-lg rounded-md bg-white">
+                <h5 class="text-lg font-bold mb-4">請提供您的反饋意見</h5>
+                <textarea id="userFeedback" class="w-full h-32 p-2 border rounded" placeholder="請輸入您的反饋..."></textarea>
+                <button id="submitFeedback" class="mt-4 bg-blue-500 hover:bg-blue-700 text-white font-bold py-2 px-4 rounded">
+                    提交
+                </button>
+                <div id="feedbackLoading" class="mt-4 text-center hidden">
+                    ⌛️ 新內容製作中...
+                </div>
+            </div>
+        </div>
+
+        <div id="segmentSummaries" class="mt-6 hidden">
+            <h3 class="text-lg font-semibold mb-2">🔗 段落摘要：</h3>
+            <div id="summaries" class="summary-content p-4 bg-white rounded shadow"></div>
+        </div>
+
+        <div class="mt-6">
+            <button id="toggleTranscript" class="btn-custom px-4 py-2 bg-gray-600 text-white rounded">顯示/隱藏轉錄文本</button>
+            <div id="transcript" class="mt-4 p-4 bg-white rounded shadow transcript-content"></div>
+        </div>
+    </div>
+
+    <script src="/static/main.js"></script>
+</body>
+</html>