Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,168 +1,42 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
import google.generativeai as genai
|
| 4 |
-
import pysrt
|
| 5 |
-
import re
|
| 6 |
import os
|
| 7 |
-
import csv
|
| 8 |
from google.colab import files
|
|
|
|
| 9 |
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
if api_key:
|
| 14 |
-
try:
|
| 15 |
-
genai.configure(api_key=api_key)
|
| 16 |
-
model = genai.GenerativeModel('gemini-1.5-flash') # 測試金鑰是否正確
|
| 17 |
-
print("API 設定完成!")
|
| 18 |
-
break
|
| 19 |
-
except Exception as e:
|
| 20 |
-
print(f"API 金鑰錯誤,請重新輸入: {e}")
|
| 21 |
-
else:
|
| 22 |
-
print("請輸入 API 金鑰")
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
# 解析 SRT 檔案
|
| 26 |
-
def parse_srt(srt_path):
|
| 27 |
-
try:
|
| 28 |
-
subs = pysrt.open(srt_path, encoding='utf-8')
|
| 29 |
-
srt_data = {}
|
| 30 |
-
for sub in subs:
|
| 31 |
-
srt_data[sub.index] = {"time": str(sub.start) + " --> " + str(sub.end), "text": sub.text}
|
| 32 |
-
return srt_data
|
| 33 |
-
except Exception as e:
|
| 34 |
-
print(f"解析字幕檔案錯誤: {e}")
|
| 35 |
-
return None
|
| 36 |
-
|
| 37 |
-
# 2. 使用者介面上傳逐字稿檔案
|
| 38 |
-
print("請上傳逐字稿檔案 (TXT 格式)")
|
| 39 |
-
uploaded_transcript_file = files.upload()
|
| 40 |
-
if uploaded_transcript_file:
|
| 41 |
-
test_transcript_file = list(uploaded_transcript_file.keys())[0]
|
| 42 |
-
else:
|
| 43 |
-
print("沒有上傳任何檔案,請重新執行")
|
| 44 |
-
exit()
|
| 45 |
-
|
| 46 |
-
# 檢查測試用逐字稿檔案是否存在
|
| 47 |
-
if not os.path.exists(test_transcript_file):
|
| 48 |
-
print(f"錯誤: 找不到逐字稿檔案: {test_transcript_file},請確認您已上傳該檔案")
|
| 49 |
-
else:
|
| 50 |
-
with open(test_transcript_file, 'r', encoding='utf-8') as f:
|
| 51 |
transcript_content = f.read()
|
| 52 |
|
| 53 |
-
|
| 54 |
-
def preprocess_transcript(transcript):
|
| 55 |
-
# 移除 SSML 標籤
|
| 56 |
-
transcript = re.sub(r'<[^>]*?>', '', transcript)
|
| 57 |
-
# 移除多個連續空格
|
| 58 |
-
transcript = re.sub(r'\s+', ' ', transcript)
|
| 59 |
-
# 移除所有換行符號
|
| 60 |
-
transcript = transcript.replace('\n','')
|
| 61 |
-
# 移除文字前後的空白
|
| 62 |
-
transcript = transcript.strip()
|
| 63 |
-
return transcript
|
| 64 |
-
# 使用預處理函數處理逐字稿
|
| 65 |
-
transcript_content = preprocess_transcript(transcript_content)
|
| 66 |
-
|
| 67 |
-
# 3. 使用者介面上傳口語稿檔案
|
| 68 |
-
print("請上傳口語稿檔案 (SRT 格式)")
|
| 69 |
-
uploaded_oral_file = files.upload()
|
| 70 |
-
if uploaded_oral_file:
|
| 71 |
-
test_srt_file = list(uploaded_oral_file.keys())[0]
|
| 72 |
-
else:
|
| 73 |
-
print("沒有上傳任何檔案,請重新執行")
|
| 74 |
-
exit()
|
| 75 |
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
print(f"錯誤: 找不到口語稿檔案: {test_srt_file},請確認您已上傳該檔案")
|
| 79 |
else:
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
prompt = (
|
| 107 |
-
"請扮演一位專業的文字編輯,比對以下字幕檔和逐字稿,並以逐字稿的內容為準,修正字幕檔中因語音轉文字造成的錯別字、漏字、多字和詞語誤用等問題,請勿修改語氣和修辭,直接輸出修正後的字幕檔。\n"
|
| 108 |
-
"接著,請將**當前批次**所有修改的內容整理成一個修改清單,並放在所有字幕內容之後。\n"
|
| 109 |
-
"請使用以下格式輸出修改清單: \n"
|
| 110 |
-
"編號: 原文: 修正後\n"
|
| 111 |
-
"請使用 `<<<分隔符號>>>` 來分隔字幕與修改清單。\n"
|
| 112 |
-
"請按照以下步驟進行:\n"
|
| 113 |
-
"1. 比對字幕檔和逐字稿。\n"
|
| 114 |
-
"2. 以逐字稿的內容為準,修正字幕檔中因語音轉文字造成的錯���字、漏字、多字和詞語誤用、以及標點符號漏缺等問題。\n"
|
| 115 |
-
"3. 必須嚴格保持原始字幕的時間戳和斷句結構,不要拆分或合併任何段落,不要新增或移除任何字幕段落,直接輸出修正後的字幕檔。\n"
|
| 116 |
-
"4. 讀取整份逐字稿時,請忽略SSML標籤,直接比對文字與標點符號,請勿修改任何語義。\n"
|
| 117 |
-
"字幕檔內容:\n"
|
| 118 |
-
f"{subtitle_content}\n\n"
|
| 119 |
-
"逐字稿內容:\n"
|
| 120 |
-
f"{transcript_content}\n\n"
|
| 121 |
-
)
|
| 122 |
-
try:
|
| 123 |
-
model = genai.GenerativeModel('gemini-1.5-flash')
|
| 124 |
-
response = model.generate_content(prompt)
|
| 125 |
-
corrected_subtitle = response.text
|
| 126 |
-
print(f"第 {i // batch_size + 1} 批次 Gemini 模型的回應:")
|
| 127 |
-
print(corrected_subtitle)
|
| 128 |
-
|
| 129 |
-
# 使用 re.split 分割字幕和報告
|
| 130 |
-
parts = re.split(r'<<<分隔符號>>>', corrected_subtitle, maxsplit=1) # 只分割一次
|
| 131 |
-
corrected_lines = parts[0].split('\n') # 取得字幕文字
|
| 132 |
-
report = ""
|
| 133 |
-
if len(parts) > 1:
|
| 134 |
-
report = parts[1] # 取得修改清單文字
|
| 135 |
-
report_lines = report.strip().split('\n')
|
| 136 |
-
all_reports.extend(report_lines)
|
| 137 |
-
|
| 138 |
-
for line in corrected_lines:
|
| 139 |
-
match = re.match(r'編號(\d+):(.*)', line)
|
| 140 |
-
if match:
|
| 141 |
-
index = int(match.group(1))
|
| 142 |
-
corrected_text = match.group(2).strip()
|
| 143 |
-
srt_data[index]['text'] = corrected_text
|
| 144 |
-
except Exception as e:
|
| 145 |
-
print(f"第 {i // batch_size + 1} 批次修正過程失敗:{e}")
|
| 146 |
-
|
| 147 |
-
# 寫回 SRT 檔案
|
| 148 |
-
new_subs = [] # 將 new_subs = [] 移到此
|
| 149 |
-
for index, data in srt_data.items():
|
| 150 |
-
time_str = data['time']
|
| 151 |
-
start_str, end_str = time_str.split(" --> ")
|
| 152 |
-
start = pysrt.srtitem.SubRipTime.from_string(start_str)
|
| 153 |
-
end = pysrt.srtitem.SubRipTime.from_string(end_str)
|
| 154 |
-
new_subs.append(pysrt.SubRipItem(index=index, start=start, end=end, text=data['text']))
|
| 155 |
-
|
| 156 |
-
# 將 new_subs 轉換為 SubRipFile 物件
|
| 157 |
-
new_srt = pysrt.SubRipFile(items=new_subs)
|
| 158 |
-
|
| 159 |
-
corrected_file_name = "corrected_subtitle.srt"
|
| 160 |
-
new_srt.save(corrected_file_name, encoding='utf-8')
|
| 161 |
-
files.download(corrected_file_name)
|
| 162 |
-
|
| 163 |
-
# 合併修改清單並下載
|
| 164 |
-
report_text = "\n".join(all_reports)
|
| 165 |
-
report_file_name = "report.txt"
|
| 166 |
-
with open(report_file_name, 'w', encoding='utf-8') as f:
|
| 167 |
-
f.write(report_text)
|
| 168 |
-
files.download(report_file_name)
|
|
|
|
| 1 |
+
# app.py
|
| 2 |
+
import gradio as gr
|
|
|
|
|
|
|
|
|
|
| 3 |
import os
|
|
|
|
| 4 |
from google.colab import files
|
| 5 |
+
from core import process_files
|
| 6 |
|
| 7 |
+
def correct_subtitle(api_key, transcript_file, oral_file, batch_size):
|
| 8 |
+
|
| 9 |
+
with open(transcript_file.name, 'r', encoding='utf-8') as f:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
transcript_content = f.read()
|
| 11 |
|
| 12 |
+
error, new_srt, all_reports = process_files(api_key, transcript_file.name, oral_file.name, batch_size)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
+
if error:
|
| 15 |
+
return error, None, None
|
|
|
|
| 16 |
else:
|
| 17 |
+
corrected_file_name = "corrected_subtitle.srt"
|
| 18 |
+
new_srt.save(corrected_file_name, encoding='utf-8')
|
| 19 |
+
report_text = "\n".join(all_reports)
|
| 20 |
+
report_file_name = "report.txt"
|
| 21 |
+
with open(report_file_name, 'w', encoding='utf-8') as f:
|
| 22 |
+
f.write(report_text)
|
| 23 |
+
return "完成",corrected_file_name, report_file_name
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
iface = gr.Interface(
|
| 27 |
+
fn=correct_subtitle,
|
| 28 |
+
inputs=[
|
| 29 |
+
gr.Textbox(lines=1, placeholder="請輸入您的 Google Gemini API 金鑰"),
|
| 30 |
+
gr.File(label="請上傳逐字稿檔案 (TXT 格式)"),
|
| 31 |
+
gr.File(label="請上傳口語稿檔案 (SRT 格式)"),
|
| 32 |
+
gr.Number(label="請輸入批次大小 (請輸入整數)")
|
| 33 |
+
],
|
| 34 |
+
outputs=[
|
| 35 |
+
gr.Textbox(label="狀態"),
|
| 36 |
+
gr.File(label="修正後的字幕檔案"),
|
| 37 |
+
gr.File(label="修改清單")
|
| 38 |
+
],
|
| 39 |
+
title="字幕修正工具"
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
iface.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|