# -*- coding: utf-8 -*-
# 財政部財政資訊中心 江信宗
import gradio as gr
import tempfile
import subprocess
from groq import Groq
from zhconv_rs import zhconv
from datetime import timedelta
import os
from pathlib import Path
def check_ffmpeg():
try:
subprocess.run(["ffmpeg", "-version"], check=True, capture_output=True, text=True)
print("FFmpeg is installed and working.")
except subprocess.CalledProcessError:
print("Error: FFmpeg is not installed or not working properly.")
except FileNotFoundError:
print("Error: FFmpeg is not installed or not in the system PATH.")
def format_time(seconds):
td = timedelta(seconds=seconds)
hours, remainder = divmod(td.seconds, 3600)
minutes, seconds = divmod(remainder, 60)
milliseconds = round(td.microseconds / 1000)
return f"{hours:02d}:{minutes:02d}:{seconds:02d},{milliseconds:03d}"
def json_to_srt(segments):
srt_lines = []
for i, item in enumerate(segments, 1):
start_time = format_time(item['start'])
end_time = format_time(item['end'])
text = zhconv(item['text'], "zh-tw")
srt_lines.append(f"{i}\n{start_time} --> {end_time}\n{text}\n")
return "\n".join(srt_lines)
def validate_and_convert(file, Language, api_key):
try:
if not api_key:
os.remove(file.name)
gr.Warning("請輸入正確的 API Key！！")
return None, None, "請輸入正確的 API Key！！", None, None
if api_key == os.getenv("SPACE_ID"):
api_key = os.getenv("YOUR_API_KEY")
else:
os.remove(file.name)
gr.Warning("請輸入正確的 API Key！！")
return None, None, "請輸入正確的 API Key！！", None, None
except Exception as e:
os.remove(file.name)
gr.Warning("請輸入正確的 API Key！！")
return None, None, f"請輸入正確的 API Key！！", None, None
gr.Info("檔案上傳完成，開始轉換......")
MAX_FILE_SIZE = 200 * 1024 * 1024 # 200MB
ALLOWED_EXTENSIONS = ['mp3', 'mp4', 'mpeg', 'm4a', 'wav', 'ogg', 'flac', 'webm', 'mpga']
if file is None:
return None, None, "沒有選擇文件", None, None
try:
file_path = Path(file.name)
if not file_path.exists():
return None, None, f"找不到上傳的檔案：{file.name}", None, None
file_extension = file_path.suffix[1:].lower()
if file_extension not in ALLOWED_EXTENSIONS:
return None, None, f"不支援的檔案類型！請上傳以下格式之一的檔案：{', '.join(ALLOWED_EXTENSIONS)}", None, None
file_size = file_path.stat().st_size
if file_size > MAX_FILE_SIZE:
return None, None, "檔案已超過200MB限制，請上傳較小的檔案。", None, None
show_info = file_size > 50 * 1024 * 1024
return convert_to_mp3(file, Language, api_key, show_info)
except Exception as e:
return None, None, f"檔案處理錯誤：{str(e)}", None, None
def convert_to_mp3(file, Language, api_key, show_info):
temp_dir = tempfile.gettempdir()
input_path = file.name
output_path = os.path.join(temp_dir, "output.mp3")
command = f"ffmpeg -i \"{input_path}\" -acodec libmp3lame -b:a 48k -y \"{output_path}\""
try:
if show_info:
gr.Info("開始轉換為音檔，請稍候......")
result = subprocess.run(command, shell=True, check=True, capture_output=True, text=True)
if os.path.exists(output_path):
file_size = os.path.getsize(output_path)
if file_size > 25 * 1024 * 1024:
return None, None, "轉譯限制：MP3檔案大小不得超過25MB！", None, None
if file_size > 5 * 1024 * 1024:
show_info = True
size_info = f"MP3 檔案大小： {file_size / 1024:.2f} KB"
transcription, transcription_text = transcribe_audio(output_path, Language, api_key, show_info)
if isinstance(transcription, list):
if show_info:
gr.Info("開始產製SRT字幕檔......")
srt_content = json_to_srt(transcription)
srt_file = create_srt_file(srt_content)
return file, output_path, size_info, transcription_text, srt_file
else:
return file, output_path, size_info, transcription, None
else:
return None, None, "轉換失敗：輸出文件不存在", None, None
except subprocess.CalledProcessError as e:
error_message = f"轉換失敗：{str(e)}"
return None, None, error_message, None, None
def summarize_article(trans_text, api_key):
try:
if not api_key:
api_key = os.getenv("YOUR_API_KEY")
client = Groq(api_key=api_key)
response = client.chat.completions.create(
model="llama-3.1-70b-versatile",
messages=[
{"role": "system", "content": """
你是一個精通繁體中文和臺灣用語的中文編輯。
使用者會提供給你一段逐字稿，請協助我檢查錯字、修正標點符號的使用，盡可能保持原意不變、保留內容細節（例如故事、提到的數字和案例）的情況下，提供優化中文排版後所有的逐字稿內容。
## 限制
提供時請分成數個段落，並替這段落下個合適的標題，並依照以下限制輸出。而且無論如何，都要提供所有的逐字稿內容，請不要擅自刪減或總結成段落！
### 中文錯字修正範例
1. 「罰還」改成「罰緩」
2. 「巧巧」改成「悄悄」
3. 「辯試」改成「辨識」
4. 「規護」改成「歸戶」
5. 「披頭」改成「劈頭」
6. 「查器」改成「查緝」
### 中文排版＆標點符號的使用原則
1. 一律使用全形符號，例如用「」做引號，而不是 “”；用 "，" 做為逗號，而不是 ","。
2. 省略號是……（兩個英文省略號），不是。。。，也不是......（六個點）
3. 中文與英文或數字之間需要增加半形空格。正確用法：「Apple 課程人數已經超過 2000 人了。」；錯誤用法：「Apple課程人數已經超過2000人了。」
4. 遇到完整的英文整句、特殊名詞，其內容使用半形標點。正確用法：「賈伯斯說過："Stay hungry, stay foolish."」；錯誤用法：「賈伯斯說過："Stay hungry，stay foolish。”」
5. 一律在中英文之間增加空格
6. 在中文與數字之間增加空格
7. 在數字與單位之間增加空格
8. 全形標點與其他字符之間不加空格。
### 其他限制
- 請注意，直接輸出結果給我，不需要有開頭招呼。
- 無論如何，都要提供所有的逐字稿內容，請不要擅自刪減或總結成段落！
## 輸出格式
**{總結後的段落重點}**
{文字段落內容}
**{總結後的段落重點}**
{文字段落內容}
"""},
{"role": "user", "content": trans_text}
],
temperature=0.2
)
return response.choices[0].message.content
except Exception as e:
return f"總結失敗：{str(e)}"
def transcribe_audio(filename, Language, api_key, show_info):
try:
if not api_key:
api_key = os.getenv("YOUR_API_KEY")
if show_info:
gr.Info("開始轉譯，請稍候......")
client = Groq(api_key=api_key)
language_dict = {"繁體中文": "zh", "English": "en", "German": "de", "Spanish": "es", "Russian": "ru", "Korean": "ko", "French": "fr", "Japanese": "ja", "Portuguese": "pt", "Turkish": "tr", "Polish": "pl", "Catalan": "ca", "Dutch": "nl", "Arabic": "ar", "Swedish": "sv", "Italian": "it", "Indonesian": "id", "Hindi": "hi", "Finnish": "fi", "Vietnamese": "vi", "Hebrew": "he", "Ukrainian": "uk", "Greek": "el", "Malay": "ms", "Czech": "cs", "Romanian": "ro", "Danish": "da", "Hungarian": "hu", "Tamil": "ta", "Norwegian": "no", "Thai": "th", "Urdu": "ur", "Croatian": "hr", "Bulgarian": "bg", "Lithuanian": "lt", "Latin": "la", "Maori": "mi", "Malayalam": "ml", "Welsh": "cy", "Slovak": "sk", "Telugu": "te", "Persian": "fa", "Latvian": "lv", "Bengali": "bn", "Serbian": "sr", "Azerbaijani": "az", "Slovenian": "sl", "Kannada": "kn", "Estonian": "et", "Macedonian": "mk", "Breton": "br", "Basque": "eu", "Icelandic": "is", "Armenian": "hy", "Nepali": "ne", "Mongolian": "mn", "Bosnian": "bs", "Kazakh": "kk", "Albanian": "sq", "Swahili": "sw", "Galician": "gl", "Marathi": "mr", "Punjabi": "pa", "Sindhi": "sd", "Gujarati": "gu", "Amharic": "am", "Yiddish": "yi", "Lao": "lo", "Uzbek": "uz", "Faroese": "fo", "Haitian creole": "ht", "Pashto": "ps", "Turkmen": "tk", "Nynorsk": "nn", "Maltese": "mt", "Sanskrit": "sa", "Luxembourgish": "lb", "Myanmar": "my", "Tibetan": "bo", "Tagalog": "tl", "Malagasy": "mg", "Assamese": "as", "Tatar": "tt", "Hawaiian": "haw", "Lingala": "ln", "Hausa": "ha", "Bashkir": "ba", "Javanese": "jw", "Sundanese": "su"}
selected_language = language_dict.get(Language)
selected_model = "distil-whisper-large-v3-en" if Language == "English" else "whisper-large-v3"
with open(filename, "rb") as file:
transcription = client.audio.transcriptions.create(
file=(filename, file.read()),
model=selected_model,
response_format="verbose_json",
language=selected_language,
temperature=0.0
)
if Language == "English":
summary = transcription.text
else:
full_text = zhconv(transcription.text, "zh-tw")
if show_info:
chunks = [full_text[i:i+1000] for i in range(0, len(full_text), 1000)]
summaries = []
for i, chunk in enumerate(chunks):
gr.Info(f"正在處理第 {i+1}/{len(chunks)} 部分...")
chunk_summary = summarize_article(chunk, api_key)
summaries.append(chunk_summary.strip())
summary = "\n\n".join(summaries)
else:
summary = summarize_article(full_text, api_key)
return transcription.segments, summary.strip()
except Exception as e:
return f"轉譯失敗：{str(e)}"
def create_srt_file(srt_content):
with tempfile.NamedTemporaryFile(mode="w", delete=False, suffix=".srt", encoding="utf-8") as temp_file:
temp_file.write(srt_content)
return temp_file.name
def clear_inputs():
return None, None, None, None, None
custom_css = """
.center-aligned {
text-align: center !important;
color: #ff4081;
text-shadow: 2px 2px 4px rgba(0,0,0,0.1);
margin-bottom: 0 !important;
}
.gr-input, .gr-box, .gr-dropdown {
border-radius: 10px !important;
border: 2px solid #ff4081 !important;
margin: 0 !important;
}
.gr-input:focus, .gr-box:focus, .gr-dropdown:focus {
border-color: #f50057 !important;
box-shadow: 0 0 0 2px rgba(245,0,87,0.2) !important;
}
.file-background {
background-color: #B7E0FF !important;
padding: 15px !important;
border-radius: 10px !important;
margin: 0 !important;
height: auto;
}
.api-background {
background-color: #FFCFB3 !important;
padding: 15px !important;
border-radius: 10px !important;
margin: 0 !important;
}
.script-background {
background-color: #FEF9D9 !important;
padding: 15px !important;
border-radius: 10px !important;
margin: 0 !important;
}
.script-background textarea {
font-size: 18px !important;
background-color: #ffffff;
border: 1px solid #f0f8ff;
border-radius: 8px;
}
.srt-background {
background-color: #FFF4B5 !important;
padding: 5px !important;
border-radius: 10px !important;
margin: 0 !important;
}
.text-background {
padding: 5px !important;
border-radius: 10px !important;
border: 2px solid #B7E0FF !important;
margin: 0 !important;
}
.clear-button {
border-radius: 10px !important;
background-color: #333333 !important;
color: white !important;
font-weight: bold !important;
transition: all 0.3s ease !important;
}
.clear-button:hover {
background-color: #000000 !important;
transform: scale(1.05);
}
"""
with gr.Blocks(theme=gr.themes.Monochrome(), css=custom_css) as demo:
gr.Markdown("""
# 📝 凊彩歐北寫 - 財政部財政資訊中心 🎵
> ### **※ 玩轉聲音魅力，開拓更多可能性，自動生成 Note-taking Record，系統布署：江信宗，LLM：Whisper large-v3。**
依據 YouTube 的服務條款（ToS），請自行明確取得 YouTube 著作權人授權後再上傳影片進行轉譯！
""", elem_classes="center-aligned")
with gr.Row():
file_input = gr.File(
label="上傳影片或音訊檔",
file_count="single",
elem_classes="file-background"
)
with gr.Column():
api_key_input = gr.Textbox(label="輸入您的 API Key", type="password", placeholder="API authentication key", elem_classes="api-background")
Language = gr.Dropdown(
choices = ["繁體中文","English","Japanese","Korean","German","French","Spanish","Arabic","Italian","Portuguese","Thai","Vietnamese","Malay","Indonesian","Hindi","Bengali","Russian"],
value="繁體中文",
label="媒體檔之音訊語言",
interactive=True,
elem_classes="api-background"
)
output_audio = gr.Audio(label="轉換後的 MP3", type="filepath", elem_classes="script-background")
with gr.Row():
srt_file_output = gr.File(label="下載 SRT 字幕檔", elem_classes="srt-background")
output_text = gr.Textbox(label="音訊檔案大小", elem_classes="script-background")
clear_button = gr.Button("清除", elem_classes="clear-button")
gr.HTML(
"""
歡迎將轉譯結果製作為財資歐北共 Podcast ，重點摘要及RAG知識問答建議使用 Google NotebookLM 更佳。
"""
)
transcription_text = gr.Markdown(label="語音轉譯結果", elem_classes="text-background")
file_input.upload(
fn=validate_and_convert,
inputs=[file_input, Language, api_key_input],
outputs=[file_input, output_audio, output_text, transcription_text, srt_file_output]
)
clear_button.click(
fn=clear_inputs,
inputs=[],
outputs=[file_input, output_audio, output_text, transcription_text, srt_file_output]
)
if __name__ == "__main__":
check_ffmpeg()
if "SPACE_ID" in os.environ:
demo.queue().launch()
else:
demo.queue().launch(share=True, show_api=False)