Skip to content

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.

fix(subtitle): atomically publish fresh caption artifacts

RudyCelekli44e6d5e11832
diff --git a/app/services/subtitle.py b/app/services/subtitle.py
index 3f3419ee..71663877 100644
--- a/app/services/subtitle.py
+++ b/app/services/subtitle.py
@@ -14,6 +14,7 @@ from loguru import logger
from app.config import config
from app.models import const
from app.utils import utils
+from app.utils.subtitle_writer import write_subtitle_file
model_size = config.whisper.get("model_size", "large-v3")
device = config.whisper.get("device", "cpu")
@@ -174,8 +175,10 @@ def create(
idx += 1
sub = "\n".join(lines) + "\n"
- with open(subtitle_file, "w", encoding="utf-8") as f:
- f.write(sub)
+ if not lines:
+ logger.warning("transcription produced no subtitle cues")
+ return
+ write_subtitle_file(subtitle_file, sub)
if log_details:
logger.info(f"subtitle file created: {subtitle_file}")
@@ -349,9 +352,11 @@ def correct(subtitle_file, video_script):
corrected = True
if corrected:
- with open(subtitle_file, "w", encoding="utf-8") as fd:
- for i, item in enumerate(new_subtitle_items):
- fd.write(f"{i + 1}\n{item[1]}\n{item[2]}\n\n")
+ content = "".join(
+ f"{i + 1}\n{item[1]}\n{item[2]}\n\n"
+ for i, item in enumerate(new_subtitle_items)
+ )
+ write_subtitle_file(subtitle_file, content)
logger.info("Subtitle corrected")
else:
logger.success("Subtitle is correct")
diff --git a/app/services/task.py b/app/services/task.py
index ebdc59dc..183e1765 100644
--- a/app/services/task.py
+++ b/app/services/task.py
@@ -35,6 +35,7 @@ from app.services import (
from app.services import upload_post
from app.services import state as sm
from app.utils import file_security, utils
+from app.utils.subtitle_writer import staged_subtitle_file
# 发布请求最长可等待数分钟,不能继续占用视频生成任务的并发名额。
@@ -589,7 +590,7 @@ def generate_subtitle(task_id, params, video_script, sub_maker, audio_file):
if not params.subtitle_enabled:
return ""
- subtitle_path = path.join(utils.task_dir(task_id), "subtitle.srt")
+ final_subtitle_path = path.join(utils.task_dir(task_id), "subtitle.srt")
subtitle_provider = config.app.get("subtitle_provider", "edge").strip().lower()
logger.info(f"\n\n## generating subtitle, provider: {subtitle_provider}")
@@ -609,40 +610,46 @@ def generate_subtitle(task_id, params, video_script, sub_maker, audio_file):
is_word_level = getattr(params, "subtitle_display_mode", "sentence") == "word_by_word"
- if subtitle_provider == "edge":
- voice.create_subtitle(
- text=video_script,
- sub_maker=sub_maker,
- subtitle_file=subtitle_path,
- word_level=is_word_level,
- )
- if not os.path.exists(subtitle_path):
- # Edge 字幕偶尔会因为时间轴与文案无法匹配而没有产出文件。这里不能
- # 自动切换到 Whisper,否则首次失败会在用户不知情的情况下下载数 GB
- # 的模型。只有显式配置 Whisper 时才允许加载模型,Edge 失败则保留
- # 无字幕视频并记录原因,避免意外的网络和磁盘开销。
- logger.warning(
- "edge subtitle generation did not produce a subtitle file; "
- "skip subtitles without falling back to whisper"
+ # A failed retry must never reuse captions from an earlier narration.
+ with staged_subtitle_file(final_subtitle_path) as subtitle_path:
+ if subtitle_provider == "edge":
+ voice.create_subtitle(
+ text=video_script,
+ sub_maker=sub_maker,
+ subtitle_file=subtitle_path,
+ word_level=is_word_level,
)
- return ""
-
- if subtitle_provider == "whisper":
- subtitle.create(
- audio_file=audio_file,
- subtitle_file=subtitle_path,
- word_level=is_word_level,
- )
- if not is_word_level:
- logger.info("\n\n## correcting subtitle")
- subtitle.correct(subtitle_file=subtitle_path, video_script=video_script)
+ if not os.path.exists(subtitle_path):
+ # Edge 字幕偶尔会因为时间轴与文案无法匹配而没有产出文件。这里不能
+ # 自动切换到 Whisper,否则首次失败会在用户不知情的情况下下载数 GB
+ # 的模型。只有显式配置 Whisper 时才允许加载模型,Edge 失败则保留
+ # 无字幕视频并记录原因,避免意外的网络和磁盘开销。
+ logger.warning(
+ "edge subtitle generation did not produce a subtitle file; "
+ "skip subtitles without falling back to whisper"
+ )
+ return ""
- subtitle_lines = subtitle.file_to_subtitles(subtitle_path)
- if not subtitle_lines:
- logger.warning(f"subtitle file is invalid: {subtitle_path}")
- return ""
+ if subtitle_provider == "whisper":
+ subtitle.create(
+ audio_file=audio_file,
+ subtitle_file=subtitle_path,
+ word_level=is_word_level,
+ )
+ if not subtitle.file_to_subtitles(subtitle_path):
+ logger.warning("whisper produced no usable subtitle cues")
+ return ""
+ if not is_word_level:
+ logger.info("\n\n## correcting subtitle")
+ subtitle.correct(subtitle_file=subtitle_path, video_script=video_script)
+
+ subtitle_lines = subtitle.file_to_subtitles(subtitle_path)
+ if not subtitle_lines:
+ logger.warning(f"subtitle file is invalid: {subtitle_path}")
+ return ""
- return subtitle_path
+ os.replace(subtitle_path, final_subtitle_path)
+ return final_subtitle_path
def get_video_materials(
diff --git a/app/services/voice.py b/app/services/voice.py
index 45aaeab9..6e0d91b4 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -30,6 +30,7 @@ from openai import OpenAI
from app.config import config
from app.utils import utils
+from app.utils.subtitle_writer import staged_subtitle_file
_DEFAULT_EDGE_TTS_TIMEOUT_SECONDS = 30.0
_SILICONFLOW_TTS_TIMEOUT_SECONDS = (10, 300) # connect, read
@@ -3219,28 +3220,23 @@ def _match_script_line(script_lines: list[str], current_text: str, sub_index: in
def _write_subtitle_items(sub_items: list[str], subtitle_file: str) -> bool:
- """
- 将已经聚合好的字幕段写入到 SRT 文件,并做一次基本可读性验证。
-
- 返回值:
- - `True`:字幕文件成功落盘且可被 moviepy 解析;
- - `False`:字幕文件写入或解析失败。
- """
+ """Publish a complete, parseable SRT without destroying earlier captions."""
try:
ensure_file_path_exists(subtitle_file)
- with open(subtitle_file, "w", encoding="utf-8") as file:
- file.write("\n".join(sub_items) + "\n")
-
- sbs = subtitles.file_to_subtitles(subtitle_file, encoding="utf-8")
- duration = max([tb for ((ta, tb), txt) in sbs]) if sbs else 0
+ with staged_subtitle_file(subtitle_file) as staged:
+ with open(staged, "w", encoding="utf-8") as file:
+ file.write("\n".join(sub_items) + "\n")
+ sbs = subtitles.file_to_subtitles(staged, encoding="utf-8")
+ if not sbs:
+ raise ValueError("subtitle output contains no cues")
+ duration = max(tb for ((ta, tb), txt) in sbs)
+ os.replace(staged, subtitle_file)
logger.info(
f"completed, subtitle file created: {subtitle_file}, duration: {duration}"
)
return True
except Exception as e:
logger.error(f"failed, error: {str(e)}")
- if os.path.exists(subtitle_file):
- os.remove(subtitle_file)
return False
diff --git a/app/utils/subtitle_writer.py b/app/utils/subtitle_writer.py
new file mode 100644
index 00000000..3743e017
--- /dev/null
+++ b/app/utils/subtitle_writer.py
@@ -0,0 +1,31 @@
+"""Stage subtitle artifacts beside their destination before publication."""
+from contextlib import contextmanager
+import os
+import tempfile
+
+from loguru import logger
+
+
+@contextmanager
+def staged_subtitle_file(destination):
+ descriptor, staged = tempfile.mkstemp(
+ prefix=".subtitle-", suffix=".srt",
+ dir=os.path.dirname(os.path.abspath(destination)),
+ )
+ os.close(descriptor)
+ try:
+ yield staged
+ finally:
+ try:
+ os.remove(staged)
+ except FileNotFoundError:
+ pass
+ except OSError as cleanup_error:
+ logger.warning(f"failed to remove temporary subtitle: {cleanup_error}")
+
+
+def write_subtitle_file(destination, content):
+ with staged_subtitle_file(destination) as staged:
+ with open(staged, "w", encoding="utf-8") as output:
+ output.write(content)
+ os.replace(staged, destination)
diff --git a/test/services/test_media_srt_publication.py b/test/services/test_media_srt_publication.py
new file mode 100644
index 00000000..80e2e1cd
--- /dev/null
+++ b/test/services/test_media_srt_publication.py
@@ -0,0 +1,87 @@
+import builtins
+from types import SimpleNamespace
+
+import pytest
+
+from app.services import subtitle, task, voice
+
+
+CUE = '1\n00:00:00,000 --> 00:00:01,000\nOld transcript\n\n'
+
+
+def test_edge_generation_does_not_reuse_previous_task_captions(tmp_path, monkeypatch):
+ destination = tmp_path / 'subtitle.srt'
+ destination.write_text(CUE, encoding='utf-8')
+ monkeypatch.setattr(task.utils, 'task_dir', lambda *args, **kwargs: str(tmp_path))
+ monkeypatch.setitem(task.config.app, 'subtitle_provider', 'edge')
+ params = SimpleNamespace(subtitle_enabled=True, subtitle_display_mode='sentence')
+ maker = SimpleNamespace(cues=[], subs=[], offset=[])
+
+ assert task.generate_subtitle('retry', params, 'Different narration.', maker, 'audio.mp3') == ''
+ assert destination.read_text(encoding='utf-8') == CUE
+ assert list(tmp_path.iterdir()) == [destination]
+
+
[email protected]('writer', ['voice', 'correction', 'whisper'])
+def test_failed_caption_write_preserves_complete_existing_srt(tmp_path, monkeypatch, writer):
+ destination = tmp_path / 'subtitle.srt'
+ destination.write_text(CUE, encoding='utf-8')
+ original = builtins.open
+
+ class PartialWrite:
+ def __init__(self, file):
+ self.file = file
+ def __enter__(self):
+ return self
+ def __exit__(self, *args):
+ self.file.close()
+ def write(self, text):
+ self.file.write(text[:5])
+ self.file.flush()
+ raise OSError('disk full during caption write')
+
+ def open_file(filename, mode='r', *args, **kwargs):
+ file = original(filename, mode, *args, **kwargs)
+ if mode == 'w' and str(filename).endswith('.srt'):
+ return PartialWrite(file)
+ return file
+
+ monkeypatch.setattr(builtins, 'open', open_file)
+ if writer == 'voice':
+ assert voice._write_subtitle_items([CUE], str(destination)) is False
+ elif writer == 'correction':
+ with pytest.raises(OSError):
+ subtitle.correct(str(destination), 'Changed transcript.')
+ else:
+ word = SimpleNamespace(word='Changed', start=0.0, end=1.0)
+ segment = SimpleNamespace(words=[word], text='Changed', start=0.0, end=1.0)
+ info = SimpleNamespace(language='en', language_probability=1.0)
+ model = SimpleNamespace(transcribe=lambda *args, **kwargs: (iter([segment]), info))
+ monkeypatch.setattr(subtitle, 'WhisperModel', object())
+ monkeypatch.setattr(subtitle, 'model', model)
+ with pytest.raises(OSError):
+ subtitle.create('local-audio.mp3', str(destination))
+ assert destination.read_text(encoding='utf-8') == CUE
+ assert list(tmp_path.iterdir()) == [destination]
+
+
+def test_subtitle_cleanup_error_does_not_hide_primary_write_failure(tmp_path, monkeypatch):
+ from app.utils import subtitle_writer
+
+ destination = tmp_path / 'subtitle.srt'
+ destination.write_text(CUE, encoding='utf-8')
+ staged = None
+
+ def cannot_cleanup(*args):
+ raise PermissionError('temporary captions are busy')
+
+ monkeypatch.setattr(subtitle_writer.os, 'remove', cannot_cleanup)
+ with pytest.raises(OSError, match='disk full during caption write'):
+ with subtitle_writer.staged_subtitle_file(str(destination)) as staged:
+ with builtins.open(staged, 'w', encoding='utf-8') as output:
+ output.write('partial captions')
+ raise OSError('disk full during caption write')
+ assert destination.read_text(encoding='utf-8') == CUE
+ assert staged != str(destination)
+ with builtins.open(staged, encoding='utf-8') as output:
+ assert output.read() == 'partial captions'
diff --git a/test/services/test_task.py b/test/services/test_task.py
index cfe71fc5..6bfc5f35 100644
--- a/test/services/test_task.py
+++ b/test/services/test_task.py
@@ -1112,13 +1112,16 @@ class TestTaskService(unittest.TestCase):
shutil.rmtree(task_dir, ignore_errors=True)
self.assertTrue(subtitle_path.endswith("subtitle.srt"))
+ created_path = create.call_args.kwargs["subtitle_file"]
+ self.assertNotEqual(created_path, subtitle_path)
+ self.assertEqual(Path(created_path).parent, Path(subtitle_path).parent)
create.assert_called_once_with(
audio_file=audio_file,
- subtitle_file=subtitle_path,
+ subtitle_file=created_path,
word_level=False,
)
correct.assert_called_once_with(
- subtitle_file=subtitle_path, video_script="Hello world."
+ subtitle_file=created_path, video_script="Hello world."
)
def test_generate_subtitle_uses_whisper_word_timing_without_correction(self):
@@ -1169,9 +1172,12 @@ class TestTaskService(unittest.TestCase):
shutil.rmtree(task_dir, ignore_errors=True)
self.assertTrue(subtitle_path.endswith("subtitle.srt"))
+ created_path = create.call_args.kwargs["subtitle_file"]
+ self.assertNotEqual(created_path, subtitle_path)
+ self.assertEqual(Path(created_path).parent, Path(subtitle_path).parent)
create.assert_called_once_with(
audio_file=audio_file,
- subtitle_file=subtitle_path,
+ subtitle_file=created_path,
word_level=True,
)
correct.assert_not_called()