利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
Python ai-video-generatorcontent-creationffmpeginstagram-reelsllmpythonshort-videosubtitlestext-to-speechtiktokvideo-automationvideo-workflowworkflow-automationyoutube-shorts
fix: keep thousand separators during subtitle splitting
Harry9c96d53571df
diff --git a/app/utils/utils.py b/app/utils/utils.pyindex c3f64933..86159da6 100644--- a/app/utils/utils.py+++ b/app/utils/utils.py@@ -185,6 +185,14 @@ def split_string_by_punctuations(s):txt += charcontinue+ if char == "," and previous_char.isdigit() and next_char.isdigit():+ # 英文数字里的千分位逗号不是断句符,例如 "1,000 years"。+ # Edge TTS 的 word boundary 通常会把这种数字整体作为连续内容返回;+ # 如果这里拆成 "1" 和 "000 years",后续字幕聚合会无法匹配脚本原文,+ # 进而错误回退到 Whisper。+ txt += char+ continue+if char not in const.PUNCTUATIONS:txt += charelse:diff --git a/test/services/test_voice.py b/test/services/test_voice.pyindex e051cbc4..7d15fc39 100644--- a/test/services/test_voice.py+++ b/test/services/test_voice.py@@ -4,7 +4,9 @@ import osimport sysimport tempfileimport time+from datetime import timedeltafrom pathlib import Path+from types import SimpleNamespacefrom unittest.mock import patch# add project root to python path@@ -313,6 +315,60 @@ class TestVoiceService(unittest.TestCase):self.assertIn("Gemini subtitle generation should work now", subtitle_content)self.assertIn("Testing multiple lines", subtitle_content)+ def test_script_split_keeps_thousand_separator_comma(self):+ """+ Edge TTS 会把 "1,000 years" 作为连续文本返回。脚本断句时不能把+ 数字中间的英文逗号当成句子边界,否则字幕聚合会出现 issue #894+ 里的 sub_items 数量少于 script_lines,并错误回退 Whisper。+ """+ text = (+ "It takes about 1,000 years for a single drop of water to finish "+ "the whole trip!"+ )++ self.assertEqual(+ utils.split_string_by_punctuations(text),+ [+ (+ "It takes about 1,000 years for a single drop of water to finish "+ "the whole trip"+ )+ ],+ )++ def test_edge_cue_aggregation_handles_thousand_separator_comma(self):+ """+ 复现 issue #894 的关键形态:Edge cues 中最后一句作为连续文本返回,+ 包含 `1,000 years`。脚本断句必须与 cues 聚合结果一致,不能把它+ 拆成两条字幕。+ """+ text = (+ "The ocean isn't just sitting stil, it moves around the world like a massive "+ "amusement park ride! Cold water at the North and South Poles sinks to the "+ "bottom because it is heavy and salty. At the same time, warm water from the "+ "sunny equator flows along the top to take its place. This creates a giant "+ "underwater conveyor belt that travels all the way around the Earth. It takes "+ "about 1,000 years for a single drop of water to finish the whole trip!"+ )+ script_lines = utils.split_string_by_punctuations(text)+ cues = []+ for index, line in enumerate(script_lines):+ # Edge 的 cue content 经常没有脚本里的空格和标点布局,这里去掉空格+ # 来模拟更严格的匹配场景。+ cues.append(+ SimpleNamespace(+ content=line.replace(" ", ""),+ start=timedelta(seconds=index),+ end=timedelta(seconds=index + 0.8),+ )+ )+ sub_maker = SimpleNamespace(cues=cues)++ sub_items = vs._build_subtitle_items_from_edge_cues(sub_maker, script_lines)++ self.assertEqual(len(sub_items), len(script_lines))+ self.assertIn("1,000 years", sub_items[-1])+if __name__ == "__main__":# python -m unittest test.services.test_voice.TestVoiceService.test_azure_tts_v1# python -m unittest test.services.test_voice.TestVoiceService.test_azure_tts_v2