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.

Preserve narration when SiliconFlow returns invalid audio

Rudy Celekli783b35975578
diff --git a/app/services/voice.py b/app/services/voice.py
index 8dd04c52..046bee0f 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -1442,6 +1442,7 @@ def siliconflow_tts(
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
for i in range(3): # 尝试3次
+ temporary_audio = None
try:
logger.info(
f"start siliconflow tts, model: {model}, voice: {voice}, try: {i + 1}"
@@ -1458,14 +1459,22 @@ def siliconflow_tts(
if not response.content:
logger.error("siliconflow tts returned empty audio")
return None
- # 保存音频文件
- with open(voice_file, "wb") as f:
+ # Decode a temporary file before publishing it. A 200 response
+ # may contain invalid audio, and must not destroy a previous
+ # successful narration at the same path.
+ ensure_file_path_exists(voice_file)
+ with tempfile.NamedTemporaryFile(
+ dir=os.path.dirname(os.path.abspath(voice_file)),
+ suffix=".mp3",
+ delete=False,
+ ) as f:
+ temporary_audio = f.name
f.write(response.content)
sub_maker = ensure_legacy_submaker_fields(SubMaker())
try:
- audio_clip = AudioFileClip(voice_file)
+ audio_clip = AudioFileClip(temporary_audio)
try:
audio_duration = audio_clip.duration
finally:
@@ -1481,14 +1490,9 @@ def siliconflow_tts(
# narration. Returning a fabricated duration lets an invalid
# file advance into the costly video pipeline.
logger.error(f"siliconflow tts returned invalid audio: {e}")
- try:
- os.remove(voice_file)
- except OSError as cleanup_error:
- logger.warning(
- f"failed to remove invalid siliconflow audio: {cleanup_error}"
- )
return None
+ os.replace(temporary_audio, voice_file)
logger.success(f"siliconflow tts succeeded: {voice_file}")
return populate_legacy_submaker_with_full_text(
sub_maker=sub_maker,
@@ -1511,6 +1515,14 @@ def siliconflow_tts(
return None
except Exception as e:
logger.error(f"siliconflow tts failed: {str(e)}")
+ finally:
+ if temporary_audio and os.path.exists(temporary_audio):
+ try:
+ os.unlink(temporary_audio)
+ except OSError as cleanup_error:
+ logger.warning(
+ f"failed to remove temporary siliconflow audio: {cleanup_error}"
+ )
return None
diff --git a/test/services/test_voice.py b/test/services/test_voice.py
index 669754c0..6655b83d 100644
--- a/test/services/test_voice.py
+++ b/test/services/test_voice.py
@@ -1725,6 +1725,57 @@ class TestElevenLabsVoice(unittest.TestCase):
self.assertIsNone(result)
post.assert_called_once()
+ def test_siliconflow_tts_invalid_audio_preserves_existing_narration(self):
+ fake_response = SimpleNamespace(status_code=200, content=b"invalid mp3")
+ with (
+ tempfile.TemporaryDirectory() as temp_dir,
+ patch.object(vs.requests, "post", return_value=fake_response),
+ patch.object(vs, "AudioFileClip", side_effect=OSError("invalid audio")),
+ patch.object(vs.config, "siliconflow", {"api_key": "test-key"}),
+ ):
+ voice_file = Path(temp_dir) / "narration.mp3"
+ voice_file.write_bytes(b"previous valid narration")
+ result = vs.siliconflow_tts(
+ text="An example narration",
+ model="FunAudioLLM/CosyVoice2-0.5B",
+ voice="FunAudioLLM/CosyVoice2-0.5B:alex",
+ voice_rate=1.0,
+ voice_file=str(voice_file),
+ )
+
+ self.assertIsNone(result)
+ self.assertEqual(voice_file.read_bytes(), b"previous valid narration")
+ self.assertEqual(list(Path(temp_dir).iterdir()), [voice_file])
+
+ def test_siliconflow_tts_publishes_only_after_audio_validation(self):
+ fake_response = SimpleNamespace(status_code=200, content=b"new mp3")
+ with tempfile.TemporaryDirectory() as temp_dir:
+ voice_file = Path(temp_dir) / "narration.mp3"
+ voice_file.write_bytes(b"previous valid narration")
+
+ def validate_audio(candidate):
+ self.assertNotEqual(Path(candidate), voice_file)
+ self.assertEqual(voice_file.read_bytes(), b"previous valid narration")
+ self.assertEqual(Path(candidate).read_bytes(), b"new mp3")
+ return SimpleNamespace(duration=2.5, close=lambda: None)
+
+ with (
+ patch.object(vs.requests, "post", return_value=fake_response),
+ patch.object(vs, "AudioFileClip", side_effect=validate_audio),
+ patch.object(vs.config, "siliconflow", {"api_key": "test-key"}),
+ ):
+ result = vs.siliconflow_tts(
+ text="An example narration",
+ model="FunAudioLLM/CosyVoice2-0.5B",
+ voice="FunAudioLLM/CosyVoice2-0.5B:alex",
+ voice_rate=1.0,
+ voice_file=str(voice_file),
+ )
+
+ self.assertIsNotNone(result)
+ self.assertEqual(voice_file.read_bytes(), b"new mp3")
+ self.assertEqual(list(Path(temp_dir).iterdir()), [voice_file])
+
def test_pause_tag_detection_and_parsing(self):
"""测试多语言停顿标签的检测、解析与清洗。"""
sample_script = (