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(tts): validate SiliconFlow responses (#1418)

Rudy Mizrahi Celeklia43b16ba79c5
diff --git a/app/services/voice.py b/app/services/voice.py
index 7e14895a..f9db0e0c 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -32,6 +32,7 @@ from app.config import config
from app.utils import utils
_DEFAULT_EDGE_TTS_TIMEOUT_SECONDS = 30.0
+_SILICONFLOW_TTS_TIMEOUT_SECONDS = (10, 300) # connect, read
_MIMO_DEFAULT_BASE_URL = "https://api.xiaomimimo.com/v1"
_MIMO_DEFAULT_TTS_MODEL = "mimo-v2.5-tts"
MINIMAX_TTS_GLOBAL_URL = "https://api.minimax.io/v1/t2a_v2"
@@ -1428,9 +1429,17 @@ def siliconflow_tts(
f"start siliconflow tts, model: {model}, voice: {voice}, try: {i + 1}"
)
- response = requests.post(url, json=payload, headers=headers)
+ response = requests.post(
+ url,
+ json=payload,
+ headers=headers,
+ timeout=_SILICONFLOW_TTS_TIMEOUT_SECONDS,
+ )
if response.status_code == 200:
+ 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:
f.write(response.content)
@@ -1443,9 +1452,24 @@ def siliconflow_tts(
audio_duration = audio_clip.duration
finally:
audio_clip.close()
+ if (
+ not isinstance(audio_duration, (int, float))
+ or not math.isfinite(audio_duration)
+ or audio_duration <= 0
+ ):
+ raise ValueError("audio duration must be positive and finite")
except Exception as e:
- logger.warning(f"Failed to read audio duration: {str(e)}")
- audio_duration = 10.0
+ # A 200 response is not proof that the bytes contain usable
+ # 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
logger.success(f"siliconflow tts succeeded: {voice_file}")
return populate_legacy_submaker_with_full_text(
diff --git a/test/services/test_voice.py b/test/services/test_voice.py
index 631dc18c..2b457096 100644
--- a/test/services/test_voice.py
+++ b/test/services/test_voice.py
@@ -1544,6 +1544,54 @@ class TestElevenLabsVoice(unittest.TestCase):
f"({expected_end_100ns} units = {audio_duration_seconds}s)",
)
+ def test_siliconflow_tts_bounds_each_network_attempt(self):
+ """A stalled speech endpoint must not block the task indefinitely."""
+ timeouts = []
+
+ def stalled_post(_url, **kwargs):
+ timeouts.append(kwargs.get("timeout"))
+ raise vs.requests.exceptions.ReadTimeout("server stalled")
+
+ with (
+ tempfile.TemporaryDirectory() as temp_dir,
+ patch.object(vs.requests, "post", side_effect=stalled_post),
+ patch.object(vs.config, "siliconflow", {"api_key": "test-key"}),
+ ):
+ voice_file = str(Path(temp_dir) / "narration.mp3")
+ 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=voice_file,
+ )
+ self.assertFalse(Path(voice_file).exists())
+
+ self.assertIsNone(result)
+ self.assertEqual(timeouts, [(10, 300)] * 3)
+
+ def test_siliconflow_tts_rejects_invalid_success_audio(self):
+ """HTTP 200 with corrupt audio must not become a fake 10-second success."""
+ 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) as post,
+ patch.object(vs, "AudioFileClip", side_effect=OSError("invalid audio")),
+ patch.object(vs.config, "siliconflow", {"api_key": "test-key"}),
+ ):
+ voice_file = str(Path(temp_dir) / "narration.mp3")
+ 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=voice_file,
+ )
+ self.assertFalse(Path(voice_file).exists())
+
+ self.assertIsNone(result)
+ post.assert_called_once()
+
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 = (