update:内存泄漏

main/xiaozhi-server/core/handle/reportHandle.py

@@ -10,7 +10,7 @@ TTS上报功能已集成到ConnectionHandler类中。
 """
 
 import time
-
+import gc
 import opuslib_next
 
 from config.manage_api_client import report as manage_report
@@ -56,41 +56,47 @@ def opus_to_wav(conn, opus_data):
     Returns:
         bytes: WAV格式的音频数据
     """
-    decoder = opuslib_next.Decoder(16000, 1)  # 16kHz, 单声道
+    decoder = None
-    pcm_data = []
+    try:
+        decoder = opuslib_next.Decoder(16000, 1)  # 16kHz, 单声道
+        pcm_data = []
 
-    for opus_packet in opus_data:
+        for opus_packet in opus_data:
-        try:
+            try:
-            pcm_frame = decoder.decode(opus_packet, 960)  # 960 samples = 60ms
+                pcm_frame = decoder.decode(opus_packet, 960)  # 960 samples = 60ms
-            pcm_data.append(pcm_frame)
+                pcm_data.append(pcm_frame)
-        except opuslib_next.OpusError as e:
+            except opuslib_next.OpusError as e:
-            conn.logger.bind(tag=TAG).error(f"Opus解码错误: {e}", exc_info=True)
+                conn.logger.bind(tag=TAG).error(f"Opus解码错误: {e}", exc_info=True)
 
-    if not pcm_data:
+        if not pcm_data:
-        raise ValueError("没有有效的PCM数据")
+            raise ValueError("没有有效的PCM数据")
 
-    # 创建WAV文件头
+        # 创建WAV文件头
-    pcm_data_bytes = b"".join(pcm_data)
+        pcm_data_bytes = b"".join(pcm_data)
-    num_samples = len(pcm_data_bytes) // 2  # 16-bit samples
+        num_samples = len(pcm_data_bytes) // 2  # 16-bit samples
 
-    # WAV文件头
+        # WAV文件头
-    wav_header = bytearray()
+        wav_header = bytearray()
-    wav_header.extend(b"RIFF")  # ChunkID
+        wav_header.extend(b"RIFF")  # ChunkID
-    wav_header.extend((36 + len(pcm_data_bytes)).to_bytes(4, "little"))  # ChunkSize
+        wav_header.extend((36 + len(pcm_data_bytes)).to_bytes(4, "little"))  # ChunkSize
-    wav_header.extend(b"WAVE")  # Format
+        wav_header.extend(b"WAVE")  # Format
-    wav_header.extend(b"fmt ")  # Subchunk1ID
+        wav_header.extend(b"fmt ")  # Subchunk1ID
-    wav_header.extend((16).to_bytes(4, "little"))  # Subchunk1Size
+        wav_header.extend((16).to_bytes(4, "little"))  # Subchunk1Size
-    wav_header.extend((1).to_bytes(2, "little"))  # AudioFormat (PCM)
+        wav_header.extend((1).to_bytes(2, "little"))  # AudioFormat (PCM)
-    wav_header.extend((1).to_bytes(2, "little"))  # NumChannels
+        wav_header.extend((1).to_bytes(2, "little"))  # NumChannels
-    wav_header.extend((16000).to_bytes(4, "little"))  # SampleRate
+        wav_header.extend((16000).to_bytes(4, "little"))  # SampleRate
-    wav_header.extend((32000).to_bytes(4, "little"))  # ByteRate
+        wav_header.extend((32000).to_bytes(4, "little"))  # ByteRate
-    wav_header.extend((2).to_bytes(2, "little"))  # BlockAlign
+        wav_header.extend((2).to_bytes(2, "little"))  # BlockAlign
-    wav_header.extend((16).to_bytes(2, "little"))  # BitsPerSample
+        wav_header.extend((16).to_bytes(2, "little"))  # BitsPerSample
-    wav_header.extend(b"data")  # Subchunk2ID
+        wav_header.extend(b"data")  # Subchunk2ID
-    wav_header.extend(len(pcm_data_bytes).to_bytes(4, "little"))  # Subchunk2Size
+        wav_header.extend(len(pcm_data_bytes).to_bytes(4, "little"))  # Subchunk2Size
 
-    # 返回完整的WAV数据
+        # 返回完整的WAV数据
-    return bytes(wav_header) + pcm_data_bytes
+        return bytes(wav_header) + pcm_data_bytes
+    finally:
+        if decoder:
+            del decoder
+            gc.collect()
 
 
 def enqueue_tts_report(conn, text, opus_data):

main/xiaozhi-server/core/providers/asr/aliyun_stream.py

@@ -5,6 +5,7 @@ import hmac
 import base64
 import hashlib
 import asyncio
+import gc
 import requests
 import websockets
 import opuslib_next
@@ -347,4 +348,7 @@ class ASRProvider(ASRProviderBase):
 
     async def close(self):
         """关闭资源"""
-        await self._cleanup()
+        await self._cleanup(None)
+        if hasattr(self, 'decoder'):
+            del self.decoder
+            gc.collect()

main/xiaozhi-server/core/providers/asr/base.py

@@ -10,6 +10,7 @@ import traceback
 import threading
 import opuslib_next
 import concurrent.futures
+import gc
 from abc import ABC, abstractmethod
 from config.logger import setup_logging
 from typing import Optional, Tuple, List
@@ -241,6 +242,7 @@ class ASRProviderBase(ABC):
     @staticmethod
     def decode_opus(opus_data: List[bytes]) -> List[bytes]:
         """将Opus音频数据解码为PCM数据"""
+        decoder = None
         try:
             decoder = opuslib_next.Decoder(16000, 1)
             pcm_data = []
@@ -265,3 +267,7 @@ class ASRProviderBase(ABC):
         except Exception as e:
             logger.bind(tag=TAG).error(f"音频解码过程发生错误: {e}")
             return []
+        finally:
+            if decoder:
+                del decoder
+                gc.collect()

main/xiaozhi-server/core/providers/asr/doubao_stream.py

@@ -4,6 +4,7 @@ import uuid
 import asyncio
 import websockets
 import opuslib_next
+import gc
 from core.providers.asr.base import ASRProviderBase
 from config.logger import setup_logging
 from core.providers.asr.dto.dto import InterfaceType
@@ -370,6 +371,17 @@ class ASRProvider(ASRProviderBase):
                 pass
             self.forward_task = None
         self.is_processing = False
+        
+        # 显式释放decoder资源
+        if hasattr(self, 'decoder') and self.decoder:
+            try:
+                del self.decoder
+                self.decoder = None
+                gc.collect()
+                logger.bind(tag=TAG).info("Doubao decoder resources released")
+            except Exception as e:
+                logger.bind(tag=TAG).error(f"Error releasing Doubao decoder: {e}")
+
         # 清理所有连接的音频缓冲区
         if hasattr(self, '_connections'):
             for conn in self._connections.values():

main/xiaozhi-server/core/providers/asr/xunfei_stream.py

@@ -5,6 +5,7 @@ import hashlib
 import asyncio
 import websockets
 import opuslib_next
+import gc
 from time import mktime
 from datetime import datetime
 from urllib.parse import urlencode
@@ -512,6 +513,17 @@ class ASRProvider(ASRProviderBase):
                 pass
             self.forward_task = None
         self.is_processing = False
+        
+        # 显式释放decoder资源
+        if hasattr(self, 'decoder') and self.decoder:
+            try:
+                del self.decoder
+                self.decoder = None
+                gc.collect()
+                logger.bind(tag=TAG).info("Xunfei decoder resources released")
+            except Exception as e:
+                logger.bind(tag=TAG).error(f"Error releasing Xunfei decoder: {e}")
+
         # 清理所有连接的音频缓冲区
         if hasattr(self, "_connections"):
             for conn in self._connections.values():

main/xiaozhi-server/core/providers/vad/silero.py

@@ -2,6 +2,7 @@ import time
 import numpy as np
 import torch
 import opuslib_next
+import gc
 from config.logger import setup_logging
 from core.providers.vad.base import VADProviderBase
 
@@ -36,6 +37,11 @@ class VADProvider(VADProviderBase):
         # 至少要多少帧才算有语音
         self.frame_window_threshold = 3
 
+    def __del__(self):
+        if hasattr(self, 'decoder'):
+            del self.decoder
+            gc.collect()
+
     def is_vad(self, conn, opus_packet):
         try:
             pcm_frame = self.decoder.decode(opus_packet, 960)

main/xiaozhi-server/core/utils/opus_encoder_utils.py

@@ -6,6 +6,7 @@ Opus编码工具类
 import logging
 import traceback
 import numpy as np
+import gc
 from opuslib_next import Encoder
 from opuslib_next import constants
 from typing import Optional, Callable, Any
@@ -128,5 +129,10 @@ class OpusEncoderUtils:
 
     def close(self):
         """关闭编码器并释放资源"""
-        # opuslib没有明确的关闭方法，Python的垃圾回收会处理
+        if hasattr(self, 'encoder') and self.encoder:
-        pass
+            try:
+                del self.encoder
+                self.encoder = None
+                gc.collect()
+            except Exception as e:
+                logging.error(f"Error releasing Opus encoder: {e}")

main/xiaozhi-server/core/utils/util.py

@@ -8,6 +8,7 @@ import requests
 import subprocess
 import numpy as np
 import opuslib_next
+import gc
 from io import BytesIO
 from core.utils import p3
 from pydub import AudioSegment
@@ -372,26 +373,30 @@ def opus_datas_to_wav_bytes(opus_datas, sample_rate=16000, channels=1):
     将opus帧列表解码为wav字节流
     """
     decoder = opuslib_next.Decoder(sample_rate, channels)
-    pcm_datas = []
+    try:
+        pcm_datas = []
 
-    frame_duration = 60  # ms
+        frame_duration = 60  # ms
-    frame_size = int(sample_rate * frame_duration / 1000)  # 960
+        frame_size = int(sample_rate * frame_duration / 1000)  # 960
 
-    for opus_frame in opus_datas:
+        for opus_frame in opus_datas:
-        # 解码为PCM（返回bytes，2字节/采样点）
+            # 解码为PCM（返回bytes，2字节/采样点）
-        pcm = decoder.decode(opus_frame, frame_size)
+            pcm = decoder.decode(opus_frame, frame_size)
-        pcm_datas.append(pcm)
+            pcm_datas.append(pcm)
 
-    pcm_bytes = b"".join(pcm_datas)
+        pcm_bytes = b"".join(pcm_datas)
 
-    # 写入wav字节流
+        # 写入wav字节流
-    wav_buffer = BytesIO()
+        wav_buffer = BytesIO()
-    with wave.open(wav_buffer, "wb") as wf:
+        with wave.open(wav_buffer, "wb") as wf:
-        wf.setnchannels(channels)
+            wf.setnchannels(channels)
-        wf.setsampwidth(2)  # 16bit
+            wf.setsampwidth(2)  # 16bit
-        wf.setframerate(sample_rate)
+            wf.setframerate(sample_rate)
-        wf.writeframes(pcm_bytes)
+            wf.writeframes(pcm_bytes)
-    return wav_buffer.getvalue()
+        return wav_buffer.getvalue()
+    finally:
+        del decoder
+        gc.collect()
 
 
 def check_vad_update(before_config, new_config):