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): avoid resubmitting accepted audio (#1516)

Rudy Mizrahi Celeklidff4a0a107eb
diff --git a/app/services/voice.py b/app/services/voice.py
index 6e0d91b4..31fd6c0c 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -1577,6 +1577,7 @@ def siliconflow_tts(
for i in range(3): # 尝试3次
temporary_audio = None
+ response_accepted = False
try:
logger.info(
f"start siliconflow tts, model: {model}, voice: {voice}, try: {i + 1}"
@@ -1590,6 +1591,7 @@ def siliconflow_tts(
)
if response.status_code == 200:
+ response_accepted = True
if not response.content:
logger.error("siliconflow tts returned empty audio")
return None
@@ -1649,6 +1651,10 @@ def siliconflow_tts(
return None
except Exception as e:
logger.error(f"siliconflow tts failed: {str(e)}")
+ if response_accepted:
+ # Retrying local processing cannot recover this accepted response
+ # and would submit another synthesis request.
+ return None
finally:
if temporary_audio and os.path.exists(temporary_audio):
try:
@@ -2492,6 +2498,7 @@ def _openai_compatible_tts(
for i in range(3):
temporary_audio = None
+ response_accepted = False
try:
logger.info(f"start {provider} tts, voice: {voice}, try: {i + 1}")
ensure_file_path_exists(voice_file)
@@ -2503,6 +2510,7 @@ def _openai_compatible_tts(
)
continue
+ response_accepted = True
if not response.content:
raise ValueError(f"{provider} returned empty audio")
@@ -2533,6 +2541,10 @@ def _openai_compatible_tts(
)
except Exception as e:
logger.error(f"{provider} tts failed: {str(e)}")
+ if response_accepted:
+ # Retrying local processing cannot recover this accepted response
+ # and would submit another synthesis request.
+ return None
finally:
if temporary_audio and os.path.exists(temporary_audio):
try:
diff --git a/test/services/test_kokoro.py b/test/services/test_kokoro.py
index f3a086a4..a0298ed2 100644
--- a/test/services/test_kokoro.py
+++ b/test/services/test_kokoro.py
@@ -118,9 +118,10 @@ def test_failed_audio_never_overwrites_output(monkeypatch, tmp_path, provider, f
) is None
assert output.read_bytes() == b"previous audio"
assert list(tmp_path.iterdir()) == [output]
- assert post.call_count == 3
+ expected_attempts = 3 if failure in {"http", "timeout"} else 1
+ assert post.call_count == expected_attempts
if failure in {"zero", "nan", "replace"}:
- assert clip.close.call_count == 3
+ clip.close.assert_called_once()
def test_transient_error_retries_successfully(monkeypatch, tmp_path):
diff --git a/test/services/test_media_accepted_audio_retries.py b/test/services/test_media_accepted_audio_retries.py
new file mode 100644
index 00000000..25d04021
--- /dev/null
+++ b/test/services/test_media_accepted_audio_retries.py
@@ -0,0 +1,56 @@
+from unittest.mock import Mock
+
+from pydub import AudioSegment
+import pytest
+import requests
+
+from app.services import voice
+
+
+def accepted_audio():
+ response = requests.Response()
+ response.status_code = 200
+ with AudioSegment.silent(duration=200).export(format='mp3') as encoded:
+ response._content = encoded.read()
+ return response
+
+
+def invoke(provider, output):
+ if provider == 'siliconflow':
+ return voice.siliconflow_tts('Hello world.', 'test-model', 'test-voice', 1.0, str(output))
+ return voice._openai_compatible_tts(
+ 'test-provider', 'http://127.0.0.1:9999/v1', '', 'test-model', 'test-voice',
+ 'Hello world.', 1.0, str(output),
+ )
+
+
[email protected]('provider', ['siliconflow', 'compatible'])
+def test_accepted_audio_publication_failure_does_not_repeat_post(tmp_path, monkeypatch, accepted_audio, provider):
+ output = tmp_path / 'voice.mp3'
+ output.write_bytes(b'previous complete narration')
+ post = Mock(return_value=accepted_audio)
+ monkeypatch.setattr(voice.requests, 'post', post)
+ monkeypatch.setitem(voice.config.siliconflow, 'api_key', 'test-local-key')
+
+ def cannot_publish(*args):
+ raise OSError('destination is unavailable')
+
+ monkeypatch.setattr(voice.os, 'replace', cannot_publish)
+ assert invoke(provider, output) is None
+ assert post.call_count == 1
+ assert output.read_bytes() == b'previous complete narration'
+ assert list(tmp_path.iterdir()) == [output]
+
+
[email protected]('provider', ['siliconflow', 'compatible'])
+def test_connect_timeout_keeps_existing_retry_and_success_policy(tmp_path, monkeypatch, accepted_audio, provider):
+ output = tmp_path / 'voice.mp3'
+ post = Mock(side_effect=[requests.ConnectTimeout('before connection'), accepted_audio])
+ monkeypatch.setattr(voice.requests, 'post', post)
+ monkeypatch.setitem(voice.config.siliconflow, 'api_key', 'test-local-key')
+
+ assert invoke(provider, output) is not None
+ assert post.call_count == 2
+ assert voice.get_audio_duration(str(output)) > 0
+ assert list(tmp_path.iterdir()) == [output]