diff --git a/main/xiaozhi-server/core/handle/helloHandle.py b/main/xiaozhi-server/core/handle/helloHandle.py index 3f4a11e3b..62d4063a8 100644 --- a/main/xiaozhi-server/core/handle/helloHandle.py +++ b/main/xiaozhi-server/core/handle/helloHandle.py @@ -8,7 +8,7 @@ from typing import TYPE_CHECKING if TYPE_CHECKING: from core.connection import ConnectionHandler from core.utils.dialogue import Message -from core.utils.util import audio_to_data +from core.utils.util import audio_to_data, audio_file_to_wav_bytes from core.providers.tts.dto.dto import SentenceType from core.utils.wakeup_word import WakeupWordsConfig from core.handle.sendAudioHandle import sendAudioMessage, send_tts_message @@ -148,8 +148,13 @@ async def wakeupWordsResponse(conn: "ConnectionHandler"): # 获取当前音色 voice = getattr(conn.tts, "voice", "default") - # 使用链接的sample_rate - wav_bytes = opus_datas_to_wav_bytes(tts_result, sample_rate=conn.sample_rate) + # 保留的 TTS 音频是文件路径而非 Opus 帧：直接转目标采样率 wav，跳过有损 Opus 往返 + if isinstance(tts_result, str): + wav_bytes = await asyncio.to_thread( + audio_file_to_wav_bytes, tts_result, conn.sample_rate + ) + else: + wav_bytes = opus_datas_to_wav_bytes(tts_result, sample_rate=conn.sample_rate) file_path = wakeup_words_config.generate_file_path(voice) with open(file_path, "wb") as f: f.write(wav_bytes) diff --git a/main/xiaozhi-server/core/utils/util.py b/main/xiaozhi-server/core/utils/util.py index 839e46a85..c5fa3fa9b 100644 --- a/main/xiaozhi-server/core/utils/util.py +++ b/main/xiaozhi-server/core/utils/util.py @@ -475,6 +475,21 @@ def opus_datas_to_wav_bytes(opus_datas, sample_rate=16000, channels=1): pass +def audio_file_to_wav_bytes(audio_file_path, sample_rate=16000): + """音频文件直接转目标采样率的 wav 字节流，不经 Opus 编解码，避免有损""" + file_type = os.path.splitext(audio_file_path)[1] + if file_type: + file_type = file_type.lstrip(".") + with _ffmpeg_path_override(): + audio = AudioSegment.from_file( + audio_file_path, format=file_type, parameters=["-nostdin"] + ) + audio = audio.set_channels(1).set_frame_rate(sample_rate).set_sample_width(2) + buffer = BytesIO() + audio.export(buffer, format="wav") + return buffer.getvalue() + + def check_vad_update(before_config, new_config): if ( new_config.get("selected_module") is None diff --git a/main/xiaozhi-server/tests/core/handle/test_wakeup_response.py b/main/xiaozhi-server/tests/core/handle/test_wakeup_response.py new file mode 100644 index 000000000..e9059cafe --- /dev/null +++ b/main/xiaozhi-server/tests/core/handle/test_wakeup_response.py @@ -0,0 +1,136 @@ +"""Wakeup caching accepts both return forms of TTSProviderBase.to_tts.""" + +import asyncio +import math +import struct +import wave +from io import BytesIO +from pathlib import Path +from types import SimpleNamespace +from unittest.mock import Mock + +import pytest + + +@pytest.fixture +def wakeup_runtime(tmp_path, monkeypatch): + # conftest supplies local configuration; no deployment config is required. + from config import settings + + monkeypatch.chdir(tmp_path) + monkeypatch.setattr(settings, "config_file_valid", True) + from core.handle import helloHandle + from core.providers.tts.base import TTSProviderBase + from core.utils.wakeup_word import WakeupWordsConfig + + class LocalTTS(TTSProviderBase): + """Replace only remote synthesis; keep the real file and Opus conversion.""" + + def __init__(self, output_dir, delete_audio_file): + super().__init__({"output_dir": str(output_dir)}, delete_audio_file) + self.voice = "test-voice" + self.filename = output_dir / "retained.wav" + buffer = BytesIO() + with wave.open(buffer, "wb") as audio: + audio.setnchannels(1) + audio.setsampwidth(2) + audio.setframerate(22050) + audio.writeframes( + b"".join( + struct.pack( + "= sample_rate + assert any(audio.readframes(audio.getnframes())) + if not delete_audio_file: + assert tts.filename.read_bytes() == tts.audio_bytes + assert not helloHandle._wakeup_response_lock.locked() + + +@pytest.mark.parametrize("tts_result", [None, []]) +def test_empty_synthesis_does_not_replace_cached_response(wakeup_runtime, tts_result): + helloHandle, wakeup_cache, _ = wakeup_runtime + wakeup_cache.generate_file_path = Mock() + wakeup_cache.update_wakeup_response = Mock() + conn = SimpleNamespace(tts=SimpleNamespace(to_tts=lambda _: tts_result)) + + asyncio.run(helloHandle.wakeupWordsResponse(conn)) + + wakeup_cache.generate_file_path.assert_not_called() + wakeup_cache.update_wakeup_response.assert_not_called() + assert not helloHandle._wakeup_response_lock.locked() + + +def test_retained_file_skips_opus_roundtrip(wakeup_runtime, tmp_path, monkeypatch): + """保留文件路径必须直转 wav，不得经过 Opus 编解码。""" + from core.utils import util + + helloHandle, wakeup_cache, LocalTTS = wakeup_runtime + + def _forbid_opus(*_args, **_kwargs): + raise AssertionError("retained file path must not run Opus encoding") + + monkeypatch.setattr(util, "pcm_to_data_stream", _forbid_opus) + tts = LocalTTS(tmp_path, False) + conn = SimpleNamespace(tts=tts, sample_rate=16000) + tts.conn = conn + + asyncio.run(helloHandle.wakeupWordsResponse(conn)) + + assert wakeup_cache.get_wakeup_response(tts.voice) is not None + + +def test_missing_retained_file_preserves_cache_and_releases_lock(wakeup_runtime): + helloHandle, wakeup_cache, _ = wakeup_runtime + wakeup_cache.generate_file_path = Mock() + wakeup_cache.update_wakeup_response = Mock() + conn = SimpleNamespace( + tts=SimpleNamespace(voice="test-voice", to_tts=lambda _: "missing.wav"), + sample_rate=16000, + ) + + with pytest.raises(FileNotFoundError): + asyncio.run(helloHandle.wakeupWordsResponse(conn)) + + wakeup_cache.generate_file_path.assert_not_called() + wakeup_cache.update_wakeup_response.assert_not_called() + assert not helloHandle._wakeup_response_lock.locked()