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): bound and atomically publish ElevenLabs speech

Rudy Celekli52c3d633da44
diff --git a/app/services/voice.py b/app/services/voice.py
index 82c15ed3..8dd04c52 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -76,6 +76,8 @@ GEMINI_TTS_VOICES = (
("Sulafat", "Warm"),
)
_MINIMAX_TTS_MAX_AUDIO_HEX_CHARS = 100 * 1024 * 1024
+_ELEVENLABS_TTS_MAX_AUDIO_BYTES = 50 * 1024 * 1024
+_ELEVENLABS_TTS_MAX_ERROR_BYTES = 4096
VOXCPM_DEFAULT_BASE_URL = "https://api.modelbest.cn/v1"
VOXCPM_DEFAULT_VOICE = "default"
VOXCPM_REFERENCE_AUDIO_MAX_UPLOAD_BYTES = 20 * 1024 * 1024
@@ -2150,41 +2152,75 @@ def elevenlabs_tts(
_NON_RETRYABLE_STATUSES = {"voice_disabled", "voice_access_denied", "unauthorized"}
for i in range(3):
+ response = None
+ temp_path = None
try:
logger.info(f"start elevenlabs tts, voice_id: {voice_id}, try: {i + 1}")
ensure_file_path_exists(voice_file)
- response = requests.post(url, json=payload, headers=headers, timeout=60)
+ response = requests.post(
+ url, json=payload, headers=headers, timeout=60, stream=True
+ )
if response.status_code != 200:
error_status = ""
+ error_bytes = bytearray()
try:
- detail = response.json().get("detail", {})
+ for chunk in response.iter_content(chunk_size=4096):
+ error_bytes.extend(
+ chunk[:_ELEVENLABS_TTS_MAX_ERROR_BYTES - len(error_bytes)]
+ )
+ if len(error_bytes) >= _ELEVENLABS_TTS_MAX_ERROR_BYTES:
+ break
+ detail = json.loads(error_bytes).get("detail", {})
if isinstance(detail, dict):
error_status = detail.get("status", "")
- except Exception:
+ except (ValueError, TypeError, AttributeError, requests.RequestException):
pass
+ error_text = error_bytes.decode("utf-8", errors="replace")[:200]
if response.status_code in _NON_RETRYABLE_CODES or error_status in _NON_RETRYABLE_STATUSES:
logger.error(
f"ElevenLabs TTS failed (non-retryable) — voice_id: {voice_id}, "
- f"status: {response.status_code}, error: {error_status or response.text[:200]}. "
+ f"status: {response.status_code}, error: {error_status or error_text}. "
"Please select a different ElevenLabs voice."
)
return None
logger.error(
- f"elevenlabs tts failed with status {response.status_code}: {response.text[:200]}"
+ f"elevenlabs tts failed with status {response.status_code}: {error_text}"
)
continue
- with open(voice_file, "wb") as f:
- f.write(response.content)
+ descriptor, temp_path = tempfile.mkstemp(
+ prefix=".elevenlabs-tts-",
+ suffix=os.path.splitext(voice_file)[1] or ".mp3",
+ dir=os.path.dirname(os.path.abspath(voice_file)),
+ )
+ audio_bytes = 0
+ with os.fdopen(descriptor, "wb") as output:
+ for chunk in response.iter_content(chunk_size=64 * 1024):
+ if not chunk:
+ continue
+ audio_bytes += len(chunk)
+ if audio_bytes > _ELEVENLABS_TTS_MAX_AUDIO_BYTES:
+ logger.error("ElevenLabs TTS audio exceeds the 50 MB limit")
+ return None
+ output.write(chunk)
+ if audio_bytes == 0:
+ logger.error("ElevenLabs TTS returned no audio data")
+ return None
- audio_clip = AudioFileClip(voice_file)
+ audio_clip = AudioFileClip(temp_path)
try:
- audio_duration = audio_clip.duration
+ audio_duration = float(audio_clip.duration)
finally:
audio_clip.close()
+ if not math.isfinite(audio_duration) or audio_duration <= 0:
+ logger.error("ElevenLabs TTS returned audio with invalid duration")
+ return None
+
+ os.replace(temp_path, voice_file)
+ temp_path = None
sub_maker = ensure_legacy_submaker_fields(SubMaker())
logger.success(f"elevenlabs tts succeeded: {voice_file}")
@@ -2203,6 +2239,18 @@ def elevenlabs_tts(
return None
except Exception as e:
logger.error(f"elevenlabs tts failed: {str(e)}")
+ if response is not None and response.status_code == 200:
+ # The provider may already have charged for a successful
+ # request. Retrying cannot repair a corrupt returned file.
+ return None
+ finally:
+ if temp_path is not None:
+ try:
+ os.remove(temp_path)
+ except OSError as exc:
+ logger.warning(f"failed to remove ElevenLabs TTS temp file: {exc}")
+ if response is not None:
+ response.close()
return None
diff --git a/test/services/test_voice.py b/test/services/test_voice.py
index d50cb9cc..669754c0 100644
--- a/test/services/test_voice.py
+++ b/test/services/test_voice.py
@@ -1015,6 +1015,12 @@ class TestVoiceService(unittest.TestCase):
content = b"fake-mp3"
text = ""
+ def iter_content(self, chunk_size):
+ yield self.content
+
+ def close(self):
+ pass
+
with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f:
out = f.name
try:
@@ -1467,6 +1473,7 @@ class TestElevenLabsVoice(unittest.TestCase):
mock_config.elevenlabs.get.return_value = "fake-api-key"
mock_post.return_value.status_code = 200
mock_post.return_value.content = b"fake-mp3-bytes"
+ mock_post.return_value.iter_content.return_value = iter((b"fake-mp3-bytes",))
mock_clip_cls.return_value.duration = 3.0
mock_clip_cls.return_value.close = lambda: None
@@ -1482,6 +1489,54 @@ class TestElevenLabsVoice(unittest.TestCase):
if os.path.exists(out_path):
os.remove(out_path)
+ def test_elevenlabs_tts_rejects_oversize_audio_without_replacing_existing_file(self):
+ """A paid 200 response can be much larger than usable speech."""
+ response = SimpleNamespace(
+ status_code=200,
+ content=b"oversize-audio",
+ iter_content=lambda chunk_size: iter((b"oversize-audio",)),
+ close=lambda: None,
+ )
+ with tempfile.TemporaryDirectory() as tmp_dir:
+ output = Path(tmp_dir) / "voice.mp3"
+ output.write_bytes(b"previous-valid-audio")
+ with (
+ patch.object(vs, "get_elevenlabs_api_key", return_value="test-key"),
+ patch.object(vs.requests, "post", return_value=response) as post,
+ patch.object(vs, "AudioFileClip") as clip,
+ patch.object(vs, "_ELEVENLABS_TTS_MAX_AUDIO_BYTES", 5, create=True),
+ ):
+ clip.return_value.duration = 1.0
+ result = vs.elevenlabs_tts("Hello", "voice-id", str(output))
+
+ self.assertIsNone(result)
+ self.assertEqual(output.read_bytes(), b"previous-valid-audio")
+ self.assertEqual(post.call_count, 1)
+ self.assertEqual(sorted(path.name for path in Path(tmp_dir).iterdir()), ["voice.mp3"])
+
+ def test_elevenlabs_tts_decode_failure_preserves_existing_file(self):
+ """A corrupt successful response must not publish a partial final MP3."""
+ response = SimpleNamespace(
+ status_code=200,
+ content=b"corrupt-audio",
+ iter_content=lambda chunk_size: iter((b"corrupt-audio",)),
+ close=lambda: None,
+ )
+ with tempfile.TemporaryDirectory() as tmp_dir:
+ output = Path(tmp_dir) / "voice.mp3"
+ output.write_bytes(b"previous-valid-audio")
+ with (
+ patch.object(vs, "get_elevenlabs_api_key", return_value="test-key"),
+ patch.object(vs.requests, "post", return_value=response) as post,
+ patch.object(vs, "AudioFileClip", side_effect=OSError("bad mp3")),
+ ):
+ result = vs.elevenlabs_tts("Hello", "voice-id", str(output))
+
+ self.assertIsNone(result)
+ self.assertEqual(output.read_bytes(), b"previous-valid-audio")
+ self.assertEqual(post.call_count, 1)
+ self.assertEqual(sorted(path.name for path in Path(tmp_dir).iterdir()), ["voice.mp3"])
+
@patch("app.services.voice.config")
def test_elevenlabs_tts_no_api_key(self, mock_config):
mock_config.elevenlabs.get.return_value = ""