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): require every pause narration chunk before publishing

Rudy Celekli064bb971e879
diff --git a/app/services/voice.py b/app/services/voice.py
index 31fd6c0c..3d5018d8 100644
--- a/app/services/voice.py
+++ b/app/services/voice.py
@@ -811,13 +811,20 @@ def _concat_audio_files(audio_files: list[str], output_file: str) -> bool:
def append_pcm(reader):
nonlocal combined_frames
+ chunk_frames = 0
while chunk := reader.readframes(8192):
+ if len(chunk) % 2:
+ raise ValueError("incomplete PCM sample in narration chunk")
combined_wave.writeframesraw(chunk)
- combined_frames += len(chunk) // 2
+ chunk_frames += len(chunk) // 2
+ if not chunk_frames or chunk_frames != reader.getnframes():
+ raise ValueError("empty or truncated narration chunk")
+ combined_frames += chunk_frames
for idx, f in enumerate(audio_files):
if not os.path.exists(f) or os.path.getsize(f) == 0:
- continue
+ logger.error(f"narration chunk is missing or empty: {f}")
+ return False
# 检查是否已经是 24000Hz 16-bit mono WAV
is_valid_pcm_wav = False
@@ -869,6 +876,7 @@ def _concat_audio_files(audio_files: list[str], output_file: str) -> bool:
return False
else:
logger.error(f"failed to decode audio chunk with ffmpeg: {res.stderr}")
+ return False
if not combined_frames:
logger.error("no valid audio samples to concatenate")
diff --git a/test/services/test_pause_concat_completeness.py b/test/services/test_pause_concat_completeness.py
new file mode 100644
index 00000000..477b1662
--- /dev/null
+++ b/test/services/test_pause_concat_completeness.py
@@ -0,0 +1,60 @@
+from pathlib import Path
+import wave
+
+import pytest
+
+from app.services import voice
+
+
+def write_pcm(target, frames=2400):
+ with wave.open(str(target), "wb") as writer:
+ writer.setnchannels(1)
+ writer.setsampwidth(2)
+ writer.setframerate(24000)
+ writer.writeframes(b"\x01\x00" * frames)
+
+
[email protected]("bad_chunk", ["missing", "empty", "invalid", "header_only", "truncated"])
[email protected]("existing", [False, True])
+def test_concat_requires_every_audio_chunk(tmp_path, bad_chunk, existing):
+ good = tmp_path / "good.wav"
+ bad = tmp_path / "bad.wav"
+ output = tmp_path / "narration.wav"
+ write_pcm(good)
+ if bad_chunk == "empty":
+ bad.write_bytes(b"")
+ elif bad_chunk == "invalid":
+ bad.write_bytes(b"not audio")
+ elif bad_chunk == "header_only":
+ write_pcm(bad, frames=0)
+ elif bad_chunk == "truncated":
+ write_pcm(bad)
+ bad.write_bytes(bad.read_bytes()[:-2])
+ if existing:
+ output.write_bytes(b"previous complete narration")
+ assert voice._concat_audio_files([str(good), str(bad)], str(output)) is False
+ assert output.read_bytes() == b"previous complete narration" if existing else not output.exists()
+
+
+def test_pause_tts_propagates_lost_chunk_without_replaying_synthesis(tmp_path, monkeypatch):
+ calls = []
+
+ def synthesize(**kwargs):
+ calls.append(kwargs["text"])
+ write_pcm(kwargs["voice_file"])
+ maker = voice.ensure_legacy_submaker_fields(voice.SubMaker())
+ return voice.populate_legacy_submaker_with_full_text(maker, kwargs["text"], 0.1)
+
+ original_concat = voice._concat_audio_files
+
+ def lose_last_chunk(files, output):
+ Path(files[-1]).unlink()
+ return original_concat(files, output)
+
+ monkeypatch.setattr(voice, "_single_tts", synthesize)
+ monkeypatch.setattr(voice, "_concat_audio_files", lose_last_chunk)
+ output = tmp_path / "narration.mp3"
+ output.write_bytes(b"previous complete narration")
+ assert voice.tts("First [pause: 0.1s] Last", "en-US-JennyNeural", 1.0, str(output)) is None
+ assert calls == ["First", "Last"]
+ assert output.read_bytes() == b"previous complete narration"