add demo:send audio to app

build.sh

@@ -1,4 +1,4 @@
-g++ -shared -fPIC \
+g++ -DBOOST_DEBUG_PYTHON -shared -fPIC \
     -I/usr/include/python3.10 -I/usr/include/python3.10/numpy -I./include \
     -L./lib -L/usr/lib/x86_64-linux-gnu \
     -DRTC_NUMPY_IMPL \

rtc_plugins.cpp

@@ -13,7 +13,11 @@ void** get_numpy_api() {
 namespace py = boost::python;
 
 int init(const char* selfUserId, const char* selfDisplayName, const char* selfRoomId, boost::python::object callback) {
-
+    if (!PyArray_API) {
+        std::cout << "PyArray_API is null in outer init" << std::endl;
+    } else {
+        std::cout << "PyArray_API is not null in outer init" << std::endl;
+    }
     RTCContext::instance().setPyCallback(callback);
     bool res = RTCContext::instance().init(selfUserId, selfDisplayName, selfRoomId);
     if (res) {
@@ -23,7 +27,11 @@ int init(const char* selfUserId, const char* selfDisplayName, const char* selfRo
     }
 }
 int initRecv(const char* destRoomId, const char* srcUserId, const int16_t destChannelIndex) {
-
+    if (!PyArray_API) {
+        std::cout << "PyArray_API is null in outer initRecv" << std::endl;
+    } else {
+        std::cout << "PyArray_API is not null in outer initRecv" << std::endl;
+    }
     bool res = RTCContext::instance().initRecv(destRoomId, srcUserId, destChannelIndex);
     if (res) {
         return 0;
@@ -31,7 +39,7 @@ int initRecv(const char* destRoomId, const char* srcUserId, const int16_t destCh
         return -1;
     }
 }
-int initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex, const uint8_t channelNum) {
+int initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex, const int16_t channelNum) {
     bool res = RTCContext::instance().initSend(srcRoomId, destRoomId, destChannelIndex, channelNum);
     if (res) {
         return 0;
@@ -39,7 +47,21 @@ int initSend(const char* srcRoomId, const char* destRoomId, const int16_t destCh
         return -1;
     }
 }
+int getSize() {
+    return RTCContext::instance().getSize();
+}
 
+namespace bp = boost::python;
+namespace np = boost::python::numpy;
+np::ndarray getNumpyData() {
+    return RTCContext::instance().getNumpyData();
+}
+bp::list getListData() {
+    return RTCContext::instance().getListData();
+}
+int16_t getDataCount() {
+    return RTCContext::instance().getDataCount();
+}
 py::object create_int16_array() {
     // 1. 定义数组维度（1维，长度为 4）
     npy_intp dims[1] = {4};
@@ -98,6 +120,65 @@ int sendCustomAudioData(int16_t destChannelIndex, py::object pD,
         return -1;
     }
 }
+RetAudioFrame getData() {
+    return RTCContext::instance().getData();
+}
+
+/*
+int sendCustomAudioData(const int16_t destChannelIndex, py::object pyData, int32_t sampleRate, uint64_t channelNum,
+                        uint64_t dataLen) {
+    try {
+        //py::object pyData = create_int16_array();
+        std::cout << "step 1" << std::endl;
+        // 1. 检查输入有效性
+        if (pyData.ptr() == nullptr) {
+            throw std::runtime_error("Input data is NULL");
+        }
+
+        std::cout << "step 2" << std::endl;
+        std::cout << "pyData ptr is:" << pyData.ptr() << std::endl;
+        if (!pyData.ptr() || !Py_IsInitialized() || !PyObject_TypeCheck(pyData.ptr(), &PyBaseObject_Type)) {
+            throw std::runtime_error("Invalid Python object");
+        }
+        std::cout << "step 2" << std::endl;
+        // 2. 检查是否是 numpy 数组
+        if (!PyArray_Check(pyData.ptr())) {
+            std::cout << "input is notnumpy" << std::endl;
+            throw std::runtime_error("Input is not a numpy array");
+        }
+
+        std::cout << "step 3" << std::endl;
+        // 3. 转换为 PyArrayObject
+        PyArrayObject* npArray = reinterpret_cast<PyArrayObject*>(pyData.ptr());
+
+        std::cout << "step 4" << std::endl;
+        // 4. 检查数据类型是否为 int16
+        if (PyArray_TYPE(npArray) != NPY_INT16) {
+            throw std::runtime_error("Array must be of type int16 (np.int16)");
+        }
+
+        std::cout << "step 5" << std::endl;
+        // 5. 检查数据是否连续
+        if (!PyArray_ISCONTIGUOUS(npArray)) {
+            throw std::runtime_error("Array must be contiguous in memory");
+        }
+        std::cout << "step 6" << std::endl;
+
+        // 6. 获取数据指针
+        void* dataPtr = PyArray_DATA(npArray);
+        if (dataPtr == nullptr) {
+            throw std::runtime_error("Invalid data pointer");
+        }
+        std::cout << "step 7" << std::endl;
+        return RTCContext::instance().sendCustomAudioData(destChannelIndex, dataPtr, sampleRate, channelNum, dataLen);
+
+    } catch (const std::exception& e) {
+        std::cout << "error:" << e.what() << std::endl;
+        return -1;
+    }
+}
+ */
+
 
 void init_numpy() {
     // 直接调用底层函数，绕过宏的问题
@@ -120,10 +201,23 @@ BOOST_PYTHON_MODULE(rtc_plugins) {
             std::cout << "set numpyApi succ:" << numpyApi[93] << std::endl;
         }
 
+        /*
+        if (!PyArray_API) {
+            std::cout << "PyArray_API is null" << std::endl;
+        } else {
+            std::cout << "PyArray_API is not null" << std::endl;
+        }
+         */
+
         py::def("init", &init);
         py::def("initRecv", &initRecv);
         py::def("initSend", &initSend);
         py::def("sendCustomAudioData", &sendCustomAudioData);
+        py::def("getSize", &getSize);
+        py::def("getData", &getData);
+        py::def("getNumpyData", &getNumpyData);
+        py::def("getListData", &getListData);
+        py::def("getDataCount", &getDataCount);
     } catch (...) {
         PyErr_SetString(PyExc_RuntimeError, "Module initialization failed");
     }

test_audio_s.py

@@ -0,0 +1,54 @@
+import rtc_plugins
+import time
+import numpy as np
+import wave
+from scipy.io import wavfile
+
+srcUserId = "srcUser12"
+destUserId = "destUser12"
+
+srcDisplayName = "srcDisplayName12"
+destDisplayName = "destDisplayName12"
+srcRoomId = "srcRoom12"
+#destRoomId = "destRoomId12"
+destRoomId = srcRoomId
+srcChannelIndex = 46
+destChannelIndex = 47
+def my_callback(shmName, dataSize, dataCount, sampleRate, numChannels, channelIndex):
+    print(f"my_callback, dataSize:{dataSize}, dataCount:{dataCount}, sampleRate:{sampleRate}, numChannels:{numChannels}, channelIndex:{channelIndex}")
+    print(f"data:{shmName}")
+    print("after my_callback_r")
+ret = rtc_plugins.init(srcUserId, srcDisplayName, srcRoomId, my_callback)
+if ret != 0:
+    print(f"init fail, ret:{ret}")
+    exit(1)
+ret = rtc_plugins.initSend(srcRoomId, destRoomId, destChannelIndex, 1)
+if ret != 0:
+    print(f"initSend fail, ret:{ret}")
+    exit(1)
+#audioData = np.array([0, 1, -1, 0], dtype=np.int16)
+sampleRate, audioData = wavfile.read("xusample1.wav")
+print(f"sampleRate:{sampleRate} HZ")
+print(f"shape:{audioData.shape}")
+print(f"type:{audioData.dtype}")
+if audioData.dtype != np.int16:
+    audioData = (audioData * 32767).astype(np.int16)
+ret = rtc_plugins.sendCustomAudioData(destChannelIndex, audioData, sampleRate, 1, len(audioData))
+if ret != 0:
+    print(f"send fail, ret:{ret}")
+ret = rtc_plugins.initRecv(srcRoomId, srcUserId, srcChannelIndex)
+if ret != 0:
+    print(f"initRecv fail, ret:{ret}")
+    exit(1)
+for i in range(100):
+    #ret = rtc_plugins.sendCustomAudioData(destChannelIndex, audioData, sampleRate, 1, len(audioData))
+    #if ret != 0:
+    #    print(f"send fail, ret:{ret}")
+
+    #size = rtc_plugins.getSize()
+    #print(f"data size:{size}")
+    #frame = rtc_plugins.getListData()
+    #print(f"get frame:{frame}")
+    #dataCount = rtc_plugins.getDataCount()
+    #print(f"data count:{dataCount}")
+    time.sleep(3)

test_send.py

@@ -17,7 +17,7 @@ ret = rtc_plugins.init(srcUserId, srcDisplayName, srcRoomId, my_callback)
 if ret != 0:
     print(f"init fail, ret:{ret}")
     exit(1)
-ret = rtc_plugins.initSend(srcRoomId, destRoomId, destChannelIndex, 1)
+ret = rtc_plugins.initSend(srcRoomId, destRoomId, destChannelIndex)
 if ret != 0:
     print(f"initSend fail, ret:{ret}")
     exit(1)

test_time_r.py

@@ -4,6 +4,7 @@ import numpy as np
 import mmap
 import os
 from ctypes import c_int16
+import struct
 
 srcUserId = "srcUser12"
 destUserId = "destUser12"
@@ -15,11 +16,12 @@ srcRoomId = "srcRoom12"
 destRoomId = srcRoomId
 srcChannelIndex = 46
 destChannelIndex = 47
-def my_callback(shmName, dataSize, dataCount, sampleRate, numChannels, channelIndex):
+def my_callback_r(shmName, dataSize, dataCount, sampleRate, numChannels, channelIndex):
-    print(f"dataSize:{dataSize}, dataCount:{dataCount}, sampleRate:{sampleRate}, numChannels:{numChannels}, channelIndex:{channelIndex}")
+    print(f"my_callback_r, dataSize:{dataSize}, dataCount:{dataCount}, sampleRate:{sampleRate}, numChannels:{numChannels}, channelIndex:{channelIndex}")
     print(f"data:{shmName}")
+    print("after my_callback_r")
 
-ret = rtc_plugins.init(destUserId, destDisplayName, destRoomId, my_callback)
+ret = rtc_plugins.init(destUserId, destDisplayName, destRoomId, my_callback_r)
 if ret != 0:
     print(f"init fail, ret:{ret}")
     exit(1)
@@ -34,10 +36,17 @@ if ret != 0:
 
 while True:
     print("recv")
-    #audioData = np.array([0, 1, -1, 0], dtype=np.int16)
+    audioData = np.array([0, 1, -1, 0], dtype=np.int16)
-    #ret = rtc_plugins.sendCustomAudioData(srcChannelIndex, audioData, 48000, 1, len(audioData))
+    ret = rtc_plugins.sendCustomAudioData(srcChannelIndex, audioData, 48000, 1, len(audioData))
-    #if ret != 0:
+    if ret != 0:
-    #    print(f"resend fail, ret:{ret}")
+        print(f"resend fail, ret:{ret}")
-    #else:
+    else:
-    #    print("resend succ")
+        print("resend succ")
-    time.sleep(30)
+    size = rtc_plugins.getSize()
+    print(f"data size:{size}")
+    #frame = rtc_plugins.getNumpyData()
+    frame = rtc_plugins.getListData()
+    print(f"get frame:{frame}")
+    dataCount = rtc_plugins.getDataCount()
+    print(f"data count:{dataCount}")
+    time.sleep(0.005)

test_time_s.py

@@ -12,8 +12,10 @@ srcRoomId = "srcRoom12"
 destRoomId = srcRoomId
 srcChannelIndex = 46
 destChannelIndex = 47
-def my_callback(npData, dataCount, sampleRate, numChannels, channelIndex):
+def my_callback(shmName, dataSize, dataCount, sampleRate, numChannels, channelIndex):
-    print(f"dataCount:{dataCount}, sampleRate:{sampleRate}, numChannels:{numChannels}, channelIndex:{channelIndex}")
+    print(f"my_callback, dataSize:{dataSize}, dataCount:{dataCount}, sampleRate:{sampleRate}, numChannels:{numChannels}, channelIndex:{channelIndex}")
+    print(f"data:{shmName}")
+    print("after my_callback_r")
 ret = rtc_plugins.init(srcUserId, srcDisplayName, srcRoomId, my_callback)
 if ret != 0:
     print(f"init fail, ret:{ret}")
@@ -34,4 +36,11 @@ for i in range(100):
     ret = rtc_plugins.sendCustomAudioData(destChannelIndex, audioData, 48000, 1, len(audioData))
     if ret != 0:
         print(f"send fail, ret:{ret}")
-    time.sleep(30)
+
+    size = rtc_plugins.getSize()
+    print(f"data size:{size}")
+    frame = rtc_plugins.getListData()
+    print(f"get frame:{frame}")
+    dataCount = rtc_plugins.getDataCount()
+    print(f"data count:{dataCount}")
+    time.sleep(3)

util/RTCContext.cpp

@@ -1,5 +1,4 @@
 #include "RTCContext.h"
-#define GIL
 
 void RTCContext::onRoom(uint32_t typeId, RTCENGINE_NAMESPACE::MRTCRoomInfo& roomInfo) {
     //std::cout << "RTCContext::onRoom():" << roomInfo.roomId << "," << roomInfo.displayName << "," << roomInfo.userId << "," << roomInfo.message;
@@ -7,30 +6,31 @@ void RTCContext::onRoom(uint32_t typeId, RTCENGINE_NAMESPACE::MRTCRoomInfo& room
     std::lock_guard<std::mutex> lock(mutex_);
     isOnRoom_ = true;
 }
-void RTCContext::onConsumer(uint32_t msgId, const char* roomId, const char* peerId,
+void RTCContext::onConsumer(uint32_t msgId, const char* roomId, const char* peerId, RTCENGINE_NAMESPACE::MRTCConsumerInfo& consumerInfo) {
-                            RTCENGINE_NAMESPACE::MRTCConsumerInfo& consumerInfo) {
+    //std::cout << "RTCContext::onConsumer()" << std::endl;
-    std::cout << "RTCContext::onConsumer():" << consumerInfo.roomId << "," << consumerInfo.displayName << ","
+    std::cout << "RTCContext::onConsumer():msgId:" << msgId << ", roomId:" << consumerInfo.roomId << ", displayName:"
-              << consumerInfo.channelIndex << std::endl;
+              << consumerInfo.displayName << ", channelIndex" << (int)consumerInfo.channelIndex;
-    if (isRecv_) {
     std::lock_guard<std::mutex> lock(mutex_);
+    isOnConsumer_ = true;
     std::cout << "registerSoundLevelListener" << std::endl;
-        int16_t ret1 = rtcEngine_->registerSoundLevelListener(mrtc::TYPE_AUDIO_SOURCE_CUSTOM, consumerInfo.roomId,
+    int16_t ret1 = rtcEngine_->registerSoundLevelListener(mrtc::TYPE_AUDIO_SOURCE_CUSTOM, roomId,
                                                           peerId, consumerInfo.channelIndex, this);
-        if (0 != ret1) {
+    if (0 != ret1)
+    {
         std::cout << "RTCContext::instance().registerSoundLevelListener() inUser failed, ret:" << ret1;
         return;
     }
 
     std::cout << "muteAudio" << std::endl;
-        int16_t ret2 = rtcEngine_->muteAudio(consumerInfo.roomId, peerId, mrtc::TYPE_AUDIO_SOURCE_CUSTOM,
+    int16_t ret2 = rtcEngine_->muteAudio(roomId, peerId, mrtc::TYPE_AUDIO_SOURCE_CUSTOM,
                                          false, consumerInfo.channelIndex);
-        if (0 != ret2) {
+    if (0 != ret2)
+    {
         std::cout << "RTCContext::instance().muteAudio() failed, ret:" << ret2;
         return;
     }
 
     std::cout << "init recv succ" << std::endl;
-    }
 }
 void RTCContext::onRender(const char* roomId, const char* peerId,
                           RTCENGINE_NAMESPACE::MRTCVideoSourceType sourceType, const RTCENGINE_NAMESPACE::MRTCVideoFrame& videoFrame) {
@@ -40,11 +40,12 @@ void RTCContext::onCallBackMessage(uint32_t msgId, const char* msg) {
 
     std::lock_guard<std::mutex> lock(mutex_);
     if (msgId == (uint32_t)mrtc::JOIN_MULTI_ROOM_SUCCESS) {
-        std::cout << "receive join multi room callback" << msgId << std::endl;
+        std::cout << "receive join multi room callback" << msgId;
         isJoinMultiRoom_ = true;
     }
 
-    std::cout << "RTCContext::onCallBackMessage(), msgId:" << msgId << ", msg:" << msg << std::endl;
+    std::cout << "RTCContext::onCallBackMessage(), msgId:" << msgId << ", msg:" << msg;
+    //std::cout << "RTCContext::onCallBackMessage()" << std::endl;
 }
 void RTCContext::onCallBackCustomData(RTCENGINE_NAMESPACE::MRTCCustomDataObject object) {
     //std::cout << "RTCContext::onCallBackCustomData(), obj:" << object.peerId << "," << object.data << "," << object.data_length;
@@ -55,24 +56,136 @@ void RTCContext::onSoundLevelUpdate(const char* roomId, const char* peerId, uint
 {
     std::cout << "RTCContext::onSoundLevelUpdate()" << std::endl;
 }
-
+/*
+void RTCContext::onAudioProcess(const char* roomId, const char* peerId,
+                                mrtc::MRTCAudioFrame& audioFrame,
+                                mrtc::MRTCAudioSourceType audioSourceType)
+{
+    namespace py = boost::python;
+
+    PyGILState_STATE gstate = PyGILState_Ensure();
+
+    try {
+        std::cout << "-----------------------------------" << std::endl;
+        std::cout << "dataCount:" << audioFrame.dataCount << std::endl;
+        std::cout << "dataCount value: " << audioFrame.dataCount
+                  << " (max: " << std::numeric_limits<npy_intp>::max() << ")" << std::endl;
+
+
+        std::cout << "onAudioProcess, numpyApi_:" << numpyApi_[93] << std::endl;
+        if (!numpyApi_ || !numpyApi_[93]) { // 93是PyArray_SimpleNew的偏移量
+            std::cout << "numpyApi_ is null in onAudioProcess" << std::endl;
+        } else {
+            std::cout << "numpyApi_ is not null in onAudioProcess:" << numpyApi_[93] << std::endl;
+        }
+        //auto numpyApi = RTCContext::numpy_api();
+        std::cout << "step1" << std::endl;
+        if (!numpyApi_) {
+            PyGILState_Release(gstate);
+            throw std::runtime_error("NumPy C-API not initialized. Call import_array() in module init");
+        }
+        std::cout << "step2" << std::endl;
+
+        using PyArray_SimpleNew_t = PyObject*(*)(int, npy_intp*, int);
+
+        void* func_ptr = numpyApi_[93];
+        std::cout << "Raw function pointer: " << func_ptr << std::endl;
+
+        auto ptmp = (PyObject*(*)(int, npy_intp*, int))numpyApi_[93];
+        std::cout << "ptmp is:" << ptmp << std::endl;
+        std::cout << "Pointer sizes:\n"
+                  << "void*: " << sizeof(void*) << "\n"
+                  << "FunctionPtr: " << sizeof(PyObject*(*)(int, npy_intp*, int)) << std::endl;
+
+        // 2. 使用memcpy避免编译器优化问题
+        PyArray_SimpleNew_t PyArray_SimpleNew;
+        static_assert(sizeof(func_ptr) == sizeof(PyArray_SimpleNew),
+                      "Pointer size mismatch");
+        std::cout << "step3" << std::endl;
+        memcpy(&PyArray_SimpleNew, &func_ptr, sizeof(func_ptr));
+
+        //auto PyArray_SimpleNew = reinterpret_cast<PyArray_SimpleNew_t>(numpyApi_[93]);
+        std::cout << "step4, PyArray_SimpleNew:" << PyArray_SimpleNew << std::endl;
+
+        // 3. 严格校验输入数据
+        if (!audioFrame.data || audioFrame.dataCount <= 0) {
+            PyGILState_Release(gstate);
+            throw std::invalid_argument("Invalid audio frame data");
+        }
+        std::cout << "step5" << std::endl;
+
+        // 4. 安全创建维度数组（带边界检查）
+        if (audioFrame.dataCount > std::numeric_limits<npy_intp>::max()) {
+            PyGILState_Release(gstate);
+            throw std::overflow_error("Audio frame size exceeds maximum limit");
+        }
+        std::cout << "step6" << std::endl;
+        npy_intp dims[1] = {static_cast<npy_intp>(audioFrame.dataCount)};
+
+        std::cout << "step7" << std::endl;
+        // 5. 创建NumPy数组（带内存保护）
+        PyObject* pyArray = nullptr;
+        pyArray = PyArray_SimpleNew(1, dims, NPY_INT16);
+        std::cout << "step8" << std::endl;
+        if (!pyArray) {
+            PyGILState_Release(gstate);
+            throw std::bad_alloc();
+        }
+        std::cout << "step9" << std::endl;
+
+        // 6. 安全拷贝数据（带对齐检查）
+        if (reinterpret_cast<uintptr_t>(audioFrame.data) % alignof(int16_t) != 0) {
+            Py_DECREF(pyArray);
+            PyGILState_Release(gstate);
+            throw std::runtime_error("Unaligned audio data pointer");
+        }
+        std::cout << "step10" << std::endl;
+        std::memcpy(PyArray_DATA(reinterpret_cast<PyArrayObject*>(pyArray)),
+                    audioFrame.data,
+                    audioFrame.dataCount * sizeof(int16_t));
+
+        std::cout << "step11" << std::endl;
+        // 7. 执行回调（带引用计数保护）
+        if (!pyCallback_.is_none()) {
+            try {
+                pyCallback_(
+                        py::handle<>(pyArray),  // 自动管理引用
+                        audioFrame.dataCount,
+                        audioFrame.sampleRate,
+                        audioFrame.numChannels,
+                        audioFrame.channelIndex
+                );
+            } catch (...) {
+                Py_DECREF(pyArray);
+                throw;  // 重新抛出异常
+            }
+        }
+        std::cout << "step12" << std::endl;
+
+        // 8. 释放资源
+        Py_DECREF(pyArray);
+        std::cout << "step13" << std::endl;
+        PyGILState_Release(gstate);
+        std::cout << "step14" << std::endl;
+
+    } catch (const std::exception& e) {
+        std::cerr << "Audio process error: " << e.what() << std::endl;
+        PyErr_Print();
+    }
+    exit(0);
+}
+ */
+/*
 void RTCContext::onAudioProcess(const char* roomId, const char* peerId,
                                 mrtc::MRTCAudioFrame& audioFrame,
                                 mrtc::MRTCAudioSourceType audioSourceType)
 {
-    namespace np = boost::python::numpy;
     namespace py = boost::python;
-    Py_Initialize();          // 初始化 Python
-    np::initialize();
     std::cout << "=== 开始音频处理 ===" << std::endl;
-    std::cout << "audioFrame:" << audioFrame.dataCount << "," << audioFrame.sampleRate << "," <<
-               audioFrame.numChannels << "," << audioFrame.channelIndex << std::endl;
 
     // 1. 获取GIL
     std::cout << "[1] 获取GIL锁..." << std::endl;
-#ifdef GIL
     PyGILState_STATE gstate = PyGILState_Ensure();
-#endif
 
     try {
         // 2. 输入参数校验
@@ -85,122 +198,285 @@ void RTCContext::onAudioProcess(const char* roomId, const char* peerId,
             throw std::invalid_argument("Invalid audio frame data");
         }
 
-        const size_t data_size = audioFrame.dataCount * sizeof(int16_t);
+        if (audioFrame.dataCount > std::numeric_limits<npy_intp>::max()) {
+            std::cout << "[ERROR] 数据长度超过最大值" << std::endl;
+            throw std::overflow_error("Audio frame size exceeds maximum limit");
+        }
 
-        std::cout << "step1" << std::endl;
+        // 3. 准备数组维度
-        namespace py = boost::python;
+        std::cout << "[3] 准备数组维度..." << std::endl;
-        namespace np = boost::python::numpy;
+        npy_intp dims[1] = {static_cast<npy_intp>(audioFrame.dataCount)};
-        npy_intp shape[1] = { static_cast<npy_intp>(audioFrame.dataCount) };
+        std::cout << "    维度设置完成: [" << dims[0] << "]" << std::endl;
-        std::cout << "step2" << std::endl;
 
-        // 7. 执行回调
+        // 4. 检查NumPy API状态
+        std::cout << "[4] 检查NumPy API状态..." << std::endl;
+        std::cout << "    numpyApi_ 地址: " << numpyApi_ << std::endl;
+        if (!numpyApi_) {
+            throw std::runtime_error("NumPy C-API not initialized");
+        }
+
+        // 5. 获取PyArray_SimpleNew函数
+        std::cout << "[5] 获取PyArray_SimpleNew函数..." << std::endl;
+        using PyArray_SimpleNew_t = PyObject*(*)(int, npy_intp*, int);
+        PyArray_SimpleNew_t PyArray_SimpleNew =
+                reinterpret_cast<PyArray_SimpleNew_t>(numpyApi_[93]);
+        std::cout << "    函数地址: " << (void*)PyArray_SimpleNew << std::endl;
+
+        std::cout << "[5.1] 验证函数指针..." << std::endl;
+        void* func_ptr = numpyApi_[93];
+        if (reinterpret_cast<uintptr_t>(func_ptr) < 0x1000) { // 检查是否为合法地址
+            std::cerr << "非法函数指针: " << func_ptr << std::endl;
+            throw std::runtime_error("Invalid PyArray_SimpleNew pointer");
+        }
+        // 6. 创建NumPy数组
+        std::cout << "[6] 创建NumPy数组..." << std::endl;
+        PyObject* pyArray = PyArray_SimpleNew(1, dims, NPY_INT16);
+        std::cout << "    数组地址: " << pyArray << std::endl;
+
+        if (!pyArray) {
+            throw std::bad_alloc();
+        }
+
+        // 7. 检查内存对齐
+        std::cout << "[7] 检查内存对齐..." << std::endl;
+        std::cout << "    音频数据地址: " << (void*)audioFrame.data
+                  << " 对齐要求: " << alignof(int16_t) << std::endl;
+
+        if (reinterpret_cast<uintptr_t>(audioFrame.data) % alignof(int16_t) != 0) {
+            Py_DECREF(pyArray);
+            throw std::runtime_error("Unaligned audio data pointer");
+        }
+
+        // 8. 拷贝数据
+        std::cout << "[8] 拷贝音频数据..." << std::endl;
+        std::cout << "    目标地址: " << PyArray_DATA((PyArrayObject*)pyArray)
+                  << " 字节数: " << audioFrame.dataCount * sizeof(int16_t) << std::endl;
+
+        std::memcpy(PyArray_DATA((PyArrayObject*)pyArray),
+                    audioFrame.data,
+                    audioFrame.dataCount * sizeof(int16_t));
+
+        // 9. 执行回调
         if (!pyCallback_.is_none()) {
-            std::cout << "[7] 准备执行Python回调..." << std::endl;
+            std::cout << "[9] 准备执行Python回调..." << std::endl;
-            // 增加引用计数防止提前释放
-            //Py_INCREF(pyCallback_.ptr());
             try {
-
-                //PyGILState_STATE gstate = PyGILState_Ensure();
-                std::cout << "data:" << audioFrame.data << std::endl;
-                std::cout << "当前线程是否持有 GIL: " << PyGILState_Check() << std::endl;
-                np::dtype dtype = np::dtype::get_builtin<int16_t>();
-                std::cout << "init dtype" << std::endl;
-                if (!Py_IsInitialized()) {
-                    std::cerr << "Python 解释器未初始化！" << std::endl;
-                    return;
-                }
-                try {
-                    py::object str_repr = py::str(dtype);
-                    std::cout << "str_repr" << std::endl;
-                    if(str_repr.ptr() != Py_None) {
-                        std::cout << "str_repr is not null" << std::endl;
-                        std::string dtype_str = py::extract<std::string>(str_repr);
-                        std::cout << "数据类型: " << dtype_str << std::endl;
-                    } else {
-                        std::cout << "数据类型: None" << std::endl;
-                    }
-                } catch (const py::error_already_set&) {
-                    std::cout<< "数据类型转换错误" << std::endl;
-                    PyErr_Clear();
-                }
-                std::cout << "数据形状: " << shape[0] << std::endl;
-                np::ndarray audioArray = np::from_data(
-                        audioFrame.data,                // 数据指针
-                        dtype,                          // 数据类型 (int16)
-                        py::make_tuple(shape[0]),       // 形状 (1D)
-                        py::make_tuple(sizeof(int16_t)), // 步长
-                        py::object()                    // 所有者（Python管理）
-                );
-                std::cout << "    数据拷贝完成" << std::endl;
                 pyCallback_(
-                        audioArray,                     // numpy 数组
+                        py::handle<>(pyArray),
-                        data_size,          // 数据大小
                         audioFrame.dataCount,
                         audioFrame.sampleRate,
                         audioFrame.numChannels,
                         audioFrame.channelIndex
                 );
-                std::cout << "    after callback" << std::endl;
-                if (PyErr_Occurred()) {
-                    PyObject *type, *value, *traceback;
-                    PyErr_Fetch(&type, &value, &traceback);
-                    if (value) {
-                        PyObject* str = PyObject_Str(value);
-                        if (str) {
-                            std::cerr << "Python Error: " << PyUnicode_AsUTF8(str) << std::endl;
-                            Py_DECREF(str);
-                        }
-                    }
-                    Py_XDECREF(type);
-                    Py_XDECREF(value);
-                    Py_XDECREF(traceback);
-                    //PyErr_Print();
-                    throw std::runtime_error("Python callback error");
-                }
                 std::cout << "    回调执行成功" << std::endl;
-
-            } catch (const py::error_already_set& e) {
-                std::cerr << "[PYTHON ERROR] ";
-                PyErr_Print();  // 自动打印到stderr
-                // 可选：获取更详细的错误信息
-                if (PyErr_Occurred()) {
-                    PyObject *type, *value, *traceback;
-                    PyErr_Fetch(&type, &value, &traceback);
-                    std::cerr << "Details: "
-                              << PyUnicode_AsUTF8(PyObject_Str(value)) << std::endl;
-                    PyErr_Restore(type, value, traceback);
-                }
-                //Py_DECREF(pyCallback_.ptr());
             } catch (...) {
                 std::cout << "[ERROR] 回调执行失败" << std::endl;
-
+                Py_DECREF(pyArray);
-                //Py_DECREF(pyCallback_.ptr());
                 throw;
             }
-            //Py_DECREF(pyCallback_.ptr());
         } else {
-            std::cout << "[7] 无回调函数设置" << std::endl;
+            std::cout << "[9] 无回调函数设置" << std::endl;
         }
 
-        // 8. 释放资源
+        // 10. 释放资源
-        std::cout << "[8] 释放共享内存资源..." << std::endl;
+        std::cout << "[10] 释放资源..." << std::endl;
-
+        Py_DECREF(pyArray);
-        std::cout << "[9] 释放GIL..." << std::endl;
+        std::cout << "[11] 释放GIL..." << std::endl;
-#ifdef GIL
         PyGILState_Release(gstate);
-#endif
         std::cout << "=== 音频处理完成 ===" << std::endl;
 
     } catch (const std::exception& e) {
         std::cout << "[EXCEPTION] 异常捕获: " << e.what() << std::endl;
-#ifdef GIL
         PyGILState_Release(gstate);
-#endif
+        PyErr_Print();
         std::cerr << "Audio process error: " << e.what() << std::endl;
     }
-#ifdef GIL
+}
-    PyGILState_Release(gstate);
+ */
-#endif
+void printTimestamp() {
+    // 获取系统当前时间点
+    auto now = std::chrono::system_clock::now();
+
+    // 转换为时间戳（秒 + 毫秒）
+    auto timestamp = std::chrono::duration_cast<std::chrono::seconds>(
+            now.time_since_epoch()).count();
+    auto milliseconds = std::chrono::duration_cast<std::chrono::milliseconds>(
+            now.time_since_epoch()).count() % 1000;
+
+    // 转换为本地时间（可读格式）
+    std::time_t time = std::chrono::system_clock::to_time_t(now);
+    std::cout << "Timestamp: " << timestamp << "." << milliseconds << std::endl;
+}
+void RTCContext::onAudioProcess(const char* roomId, const char* peerId,
+                                mrtc::MRTCAudioFrame& audioFrame,
+                                mrtc::MRTCAudioSourceType audioSourceType)
+{
+    //namespace py = boost::python;
+    //std::cout << "=== 开始音频处理(共享内存版) ===" << std::endl;
+    //std::cout << "audioFrame:" << audioFrame.dataCount << "," << audioFrame.sampleRate << "," <<
+    //           audioFrame.numChannels << "," << audioFrame.channelIndex << std::endl;
+
+    //printTimestamp();
+    setData(audioFrame);
+    // 1. 获取GIL
+    //std::cout << "[1] 获取GIL锁..." << std::endl;
+    ////PyGILState_STATE gstate = PyGILState_Ensure();
+
+    //try {
+    //    // 2. 输入参数校验
+    //    std::cout << "[2] 检查输入参数..." << std::endl;
+    //    std::cout << "    dataCount: " << audioFrame.dataCount
+    //              << " (max: " << std::numeric_limits<npy_intp>::max() << ")" << std::endl;
+
+    //    if (!audioFrame.data || audioFrame.dataCount <= 0) {
+    //        std::cout << "[ERROR] 无效音频数据指针或长度" << std::endl;
+    //        throw std::invalid_argument("Invalid audio frame data");
+    //    }
+
+    //    const size_t data_size = audioFrame.dataCount * sizeof(int16_t);
+
+    //    // 3. 创建共享内存
+    //    std::cout << "[3] 创建共享内存..." << std::endl;
+    //    char shm_name[32];
+    //    //snprintf(shm_name, sizeof(shm_name), "/audio_shm_%d", getpid());
+    //    snprintf(shm_name, sizeof(shm_name), "/audio_shm_test");
+
+    //    int fd = shm_open(shm_name, O_CREAT | O_RDWR, 0666);
+    //    if (fd == -1) {
+    //        std::cout << "[ERROR] shm_open失败: " << strerror(errno) << std::endl;
+    //        throw std::runtime_error("Failed to create shared memory");
+    //    }
+    //    std::cout << "    共享内存fd: " << fd << " 名称: " << shm_name << std::endl;
+
+    //    // 4. 设置共享内存大小
+    //    std::cout << "[4] 设置共享内存大小..." << std::endl;
+    //    if (ftruncate(fd, data_size) == -1) {
+    //        close(fd);
+    //        std::cout << "[ERROR] ftruncate失败: " << strerror(errno) << std::endl;
+    //        throw std::runtime_error("Failed to resize shared memory");
+    //    }
+    //    std::cout << "    内存大小: " << data_size << " bytes" << std::endl;
+
+    //    // 5. 内存映射
+    //    std::cout << "[5] 内存映射..." << std::endl;
+    //    void* ptr = mmap(NULL, data_size, PROT_WRITE, MAP_SHARED, fd, 0);
+    //    if (ptr == MAP_FAILED) {
+    //        close(fd);
+    //        std::cout << "[ERROR] mmap失败: " << strerror(errno) << std::endl;
+    //        throw std::runtime_error("Failed to map shared memory");
+    //    }
+    //    std::cout << "    映射地址: " << ptr << std::endl;
+
+    //    namespace py = boost::python;
+    //    namespace np = boost::python::numpy;
+    //    // 6. 拷贝数据到共享内存
+    //    std::cout << "[6] 拷贝音频数据到共享内存..." << std::endl;
+    //    memcpy(ptr, audioFrame.data, data_size);
+    //    std::cout << "step1" << std::endl;
+    //    /*
+    //    npy_intp shape[1] = { static_cast<npy_intp>(audioFrame.dataCount) };
+    //    std::cout << "step2" << std::endl;
+    //    np::dtype dtype = np::dtype::get_builtin<int16_t>();
+    //    std::cout << "step3" << std::endl;
+    //    np::ndarray audioArray = np::from_data(
+    //            audioFrame.data,                // 数据指针
+    //            dtype,                          // 数据类型 (int16)
+    //            py::make_tuple(shape[0]),       // 形状 (1D)
+    //            py::make_tuple(sizeof(int16_t)), // 步长
+    //            py::object()                    // 所有者（Python管理）
+    //    );
+    //     */
+    //    std::cout << "    数据拷贝完成" << std::endl;
+
+    //    // 7. 执行回调
+    //    //if (!pyCallback_.is_none()) {
+    //    //    std::cout << "[7] 准备执行Python回调..." << std::endl;
+    //    //    // 增加引用计数防止提前释放
+    //    //    Py_INCREF(pyCallback_.ptr());
+    //    //    try {
+    //    //        std::cout << "    pyCallback_ type: " << Py_TYPE(pyCallback_.ptr())->tp_name << std::endl;
+    //    //        PyObject* repr = PyObject_Repr(pyCallback_.ptr());
+    //    //        if (repr) {
+    //    //            std::cout << "    pyCallback_ repr: " << PyUnicode_AsUTF8(repr) << std::endl;
+    //    //            Py_DECREF(repr);  // 必须手动释放
+    //    //        }
+    //    //        // 传递共享内存信息
+    //    //        pyCallback_(
+    //    //                py::str(shm_name),  // 共享内存名称
+    //    //                data_size,          // 数据大小
+    //    //                audioFrame.dataCount,
+    //    //                audioFrame.sampleRate,
+    //    //                audioFrame.numChannels,
+    //    //                audioFrame.channelIndex
+    //    //        );
+    //    //        /*
+    //    //        pyCallback_(
+    //    //                audioArray,                     // numpy 数组
+    //    //                data_size,          // 数据大小
+    //    //                audioFrame.dataCount,
+    //    //                audioFrame.sampleRate,
+    //    //                audioFrame.numChannels,
+    //    //                audioFrame.channelIndex
+    //    //        );
+    //    //         */
+    //    //        std::cout << "    after callback" << std::endl;
+    //    //        if (PyErr_Occurred()) {
+    //    //            PyObject *type, *value, *traceback;
+    //    //            PyErr_Fetch(&type, &value, &traceback);
+    //    //            if (value) {
+    //    //                PyObject* str = PyObject_Str(value);
+    //    //                if (str) {
+    //    //                    std::cerr << "Python Error: " << PyUnicode_AsUTF8(str) << std::endl;
+    //    //                    Py_DECREF(str);
+    //    //                }
+    //    //            }
+    //    //            Py_XDECREF(type);
+    //    //            Py_XDECREF(value);
+    //    //            Py_XDECREF(traceback);
+    //    //            //PyErr_Print();
+    //    //            throw std::runtime_error("Python callback error");
+    //    //        }
+    //    //        std::cout << "    回调执行成功" << std::endl;
+
+    //    //    } catch (const py::error_already_set& e) {
+    //    //        std::cerr << "[PYTHON ERROR] ";
+    //    //        PyErr_Print();  // 自动打印到stderr
+    //    //        // 可选：获取更详细的错误信息
+    //    //        if (PyErr_Occurred()) {
+    //    //            PyObject *type, *value, *traceback;
+    //    //            PyErr_Fetch(&type, &value, &traceback);
+    //    //            std::cerr << "Details: "
+    //    //                      << PyUnicode_AsUTF8(PyObject_Str(value)) << std::endl;
+    //    //            PyErr_Restore(type, value, traceback);
+    //    //        }
+    //    //        Py_DECREF(pyCallback_.ptr());
+    //    //    } catch (...) {
+    //    //        std::cout << "[ERROR] 回调执行失败" << std::endl;
+    //    //        munmap(ptr, data_size);
+    //    //        close(fd);
+    //    //        shm_unlink(shm_name);
+    //    //        Py_DECREF(pyCallback_.ptr());
+    //    //        throw;
+    //    //    }
+    //    //    Py_DECREF(pyCallback_.ptr());
+    //    //} else {
+    //    //    std::cout << "[7] 无回调函数设置" << std::endl;
+    //    //}
+
+    //    // 8. 释放资源
+    //    std::cout << "[8] 释放共享内存资源..." << std::endl;
+    //    munmap(ptr, data_size);
+    //    close(fd);
+    //    shm_unlink(shm_name);
+
+    //    std::cout << "[9] 释放GIL..." << std::endl;
+    //    //PyGILState_Release(gstate);
+    //    std::cout << "=== 音频处理完成 ===" << std::endl;
+
+    //} catch (const std::exception& e) {
+    //    std::cout << "[EXCEPTION] 异常捕获: " << e.what() << std::endl;
+    //    //PyGILState_Release(gstate);
+    //    std::cerr << "Audio process error: " << e.what() << std::endl;
+    //}
 }
 
 void RTCContext::onProducer(uint32_t msgId, mrtc::MRTCProducerInfo& info)
@@ -261,20 +537,51 @@ bool RTCContext::init(const char* selfUserId, const char* selfDisplayName, const
         std::cout << "RTCContext::instance().registerListener() failed" << std::endl;
         return false;
     }
-    //namespace py = boost::python;
+    namespace py = boost::python;
-    //namespace np = boost::python::numpy;
+    namespace np = boost::python::numpy;
-    //Py_Initialize();          // 初始化 Python
+    Py_Initialize();          // 初始化 Python
-    //np::initialize();
+    np::initialize();
 
     return true;
 }
 bool RTCContext::initRecv(const char* destRoomId, const char* srcUserId, const int16_t destChannelIndex)
 {
-    isRecv_ = true;
+    std::cout << "initRecv, numpyApi_:" << numpyApi_[93] << std::endl;
+    if (!numpyApi_ || !numpyApi_[93]) { // 93是PyArray_SimpleNew的偏移量
+        std::cout << "numpyApi_ is null in initRecv" << std::endl;
+    } else {
+        std::cout << "numpyApi_ is not null in initRecv" << std::endl;
+    }
+    while (!isOnConsumer_)
+    {
+        std::cout << "wait for OnConsumer" << std::endl;
+        sleep(3);
+    }
+    /*
+    std::cout << "registerSoundLevelListener" << std::endl;
+    int16_t ret1 = rtcEngine_->registerSoundLevelListener(mrtc::TYPE_AUDIO_SOURCE_CUSTOM, destRoomId,
+                                                          srcUserId, destChannelIndex, this);
+    if (0 != ret1)
+    {
+        std::cout << "RTCContext::instance().registerSoundLevelListener() inUser failed, ret:" << ret1;
+        return false;
+    }
+
+    std::cout << "muteAudio" << std::endl;
+    int16_t ret2 = rtcEngine_->muteAudio(destRoomId, srcUserId, mrtc::TYPE_AUDIO_SOURCE_CUSTOM, false, destChannelIndex);
+    if (0 != ret2)
+    {
+        std::cout << "RTCContext::instance().muteAudio() failed, ret:" << ret2;
+        return false;
+    }
+
+    std::cout << "init recv succ" << std::endl;
+     */
     return true;
 
 }
-bool RTCContext::initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex, uint8_t channelNum)
+bool RTCContext::initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex,
+                          const uint8_t channelNum)
 {
     while (!isOnRoom_)
     {
@@ -295,11 +602,11 @@ bool RTCContext::initSend(const char* srcRoomId, const char* destRoomId, const i
     }
 
     mrtc::MRTCAudioOption option;
-    option.channel = channelNum;
     if (std::string(srcRoomId) != std::string(destRoomId)) {
         strcpy(option.dstRoomId, destRoomId);
     }
     option.channelIndex = destChannelIndex;
+    option.channel = channelNum;
     std::cout << "startCustomAudio" << std::endl;
     int16_t ret2 = rtcEngine_->startCustomAudio(option);
     if (ret2 != 0)
@@ -310,16 +617,12 @@ bool RTCContext::initSend(const char* srcRoomId, const char* destRoomId, const i
     std::cout << "init send succ" << std::endl;
     return true;
 }
-bool RTCContext::initGIL() {
-    isGIL_ = true;
-}
 
 void RTCContext::destorySend(const int16_t selfChannelIndex)
 {
     rtcEngine_->stopCustomAudio(selfChannelIndex);
 }
-int16_t RTCContext::sendAudioData(uint8_t channelIndex, const void* pData, int32_t nSampleRate, uint64_t nNumberOfChannels,
+int16_t RTCContext::sendAudioData(uint8_t channelIndex, const void* pData, int32_t nSampleRate, uint64_t nNumberOfChannels, uint64_t dataLength)
-                                  uint64_t dataLength)
 {
     std::lock_guard<std::mutex> lock(mutex_);
     if (pData_)
@@ -341,8 +644,7 @@ int16_t RTCContext::sendCustomAudioData(const int16_t channelIndex, void* custom
         return -1;
     }
     std::cout << "customData addr is:" << customData << std::endl;
-    return rtcEngine_->sendCustomAudioData(channelIndex, customData, sampleRate,
+    return rtcEngine_->sendCustomAudioData(channelIndex, customData, sampleRate, channelNum, dataLen);
-                                           channelNum, dataLen);
 }
 mrtc::IMRTCEngine* RTCContext::getRtcEngine() const
 {
@@ -368,3 +670,87 @@ void RTCContext::setNumpyApi(void **numpyApi) {
     numpyApi_ = numpyApi;
     std::cout << "setNupyApi, numpyApi_:" << numpyApi_[93] << std::endl;
 }
+void RTCContext::setData(const mrtc::MRTCAudioFrame& frame) {
+    std::lock_guard<std::mutex> lock(dataMutex_);
+    if (dataSize_ == totalSize_) {
+        bottom_ = (bottom_ + 1) % totalSize_;
+        dataSize_--;
+    }
+    RetAudioFrame newFrame;
+    newFrame.dataCount = frame.dataCount;
+    newFrame.sampleRate = frame.sampleRate;
+    newFrame.numChannels = frame.numChannels;
+    newFrame.channelIndex = frame.channelIndex;
+    newFrame.data = std::make_unique<int16_t[]>(frame.dataCount);
+    std::memcpy(newFrame.data.get(), frame.data, frame.dataCount* sizeof(int16_t));
+    data_[head_] = std::move(newFrame);
+    head_ = (head_ + 1) % totalSize_;
+    dataSize_++;
+
+}
+
+RetAudioFrame RTCContext::getData() {
+    //std::lock_guard<std::mutex> lock(dataMutex_);
+    if (dataSize_ > 0) {
+        RetAudioFrame frame = std::move(data_[bottom_]); // 移动而非拷贝
+        bottom_ = (bottom_ + 1) % totalSize_;
+        dataSize_--;
+        return frame; // 返回值优化(RVO)会生效
+    }
+    return {}; // 返回空对象
+}
+namespace bp = boost::python;
+namespace np = boost::python::numpy;
+np::ndarray RTCContext::getNumpyData() {
+    std::cout << "step1" << std::endl;
+    std::lock_guard<std::mutex> lock(dataMutex_);
+    RetAudioFrame frame = getData();
+    std::cout << "step2" << std::endl;
+    int16_t* dataPtr = frame.data.get();  // 你的数据指针
+    std::cout << "step3" << std::endl;
+    size_t length = frame.dataCount;    // 数据长度
+    std::cout << "step4" << std::endl;
+
+    PyGILState_STATE gstate = PyGILState_Ensure();
+    np::ndarray result = np::empty(bp::make_tuple(length), np::dtype::get_builtin<int16_t>());
+    try {
+        if (!dataPtr || length == 0) {
+            result = np::zeros(bp::make_tuple(0), np::dtype::get_builtin<int16_t>());
+        } else {
+            result = np::empty(bp::make_tuple(length), np::dtype::get_builtin<int16_t>());
+            std::memcpy(result.get_data(), dataPtr, length * sizeof(int16_t));
+        }
+    } catch (...) {
+        PyGILState_Release(gstate);  // 异常时释放GIL
+        throw;
+    }
+    PyGILState_Release(gstate);
+    return result;
+
+}
+bp::list RTCContext::getListData() {
+    std::cout << "step1" << std::endl;
+    std::lock_guard<std::mutex> lock(dataMutex_);
+    RetAudioFrame frame = getData();
+    std::cout << "step2" << std::endl;
+    int16_t* dataPtr = frame.data.get();  // 你的数据指针
+    std::cout << "step3" << std::endl;
+    size_t length = frame.dataCount;    // 数据长度
+    std::cout << "step4" << std::endl;
+    bp::list result;
+    if (dataPtr && length > 0) {
+        for (size_t i = 0; i < length; ++i) {
+            result.append(dataPtr[i]);  // 逐个元素添加（值传递）
+        }
+    }
+    return result;
+}
+int16_t RTCContext::getDataCount() {
+    std::lock_guard<std::mutex> lock(dataMutex_);
+    RetAudioFrame frame = getData();
+    return frame.dataCount;
+}
+int16_t RTCContext::getSize() {
+    std::lock_guard<std::mutex> lock(dataMutex_);
+    return dataSize_;
+}

util/RTCContext.h

@@ -31,10 +31,22 @@
 // 必须声明外部变量（关键！）
 //#define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION
 
+namespace bp = boost::python;
 namespace fs = std::filesystem;
+namespace np = boost::python::numpy;
 #define ENV_PRODUCT
 //#define SEND_MODE
 
+// 音频数据帧
+struct RetAudioFrame final
+{
+    std::unique_ptr<int16_t[]> data;
+    int dataCount = 0;
+    int sampleRate = 48000;
+    int numChannels = 1;
+    int channelIndex = 0;
+};
+
 class RTCContext :
         public RTCENGINE_NAMESPACE::IMRTCRoomCallBack,
         public RTCENGINE_NAMESPACE::IMRTCConsumerCallBack,
@@ -84,8 +96,13 @@ public:
     mrtc::IMRTCEngine* getRtcEngine() const;
     bool init(const char* selfUserId, const char* selfDisplayName, const char* selfRoomId);
     bool initRecv(const char* destRoomId, const char* srcUserId, const int16_t destChannelIndex);
-    bool initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex, uint8_t channelNum);
+    bool initSend(const char* srcRoomId, const char* destRoomId, const int16_t destChannelIndex, const uint8_t channelNum);
-    bool initGIL();
+    int16_t getSize();
+    void setData(const mrtc::MRTCAudioFrame& frame);
+    RetAudioFrame getData();
+    np::ndarray getNumpyData();
+    bp::list getListData();
+    int16_t getDataCount();
 
     void* getpData() const;
     void setpData(void* pData);
@@ -100,17 +117,23 @@ public:
 private:
     RTCContext()
     {
+        data_.resize(totalSize_);
     }
     mutable std::mutex mutex_;
     mrtc::IMRTCEngine * rtcEngine_ = nullptr;
     void* pData_ = nullptr;
     bool isOnRoom_ = false;
-    bool isRecv_ = false;
+    bool isOnConsumer_ = false;
     bool isJoinMultiRoom_ = false;
     bool isMultiRoom_ = false;
-    bool isGIL_ = false;
     boost::python::object pyCallback_;
     void ** numpyApi_;
+    std::vector<RetAudioFrame> data_;
+    mutable std::mutex dataMutex_;
+    const int16_t totalSize_ = 100;
+    int16_t dataSize_ = 0;
+    int16_t bottom_ = 0;
+    int16_t head_= 0;
     void onRoom(uint32_t typeId, RTCENGINE_NAMESPACE::MRTCRoomInfo& roomInfo);
     void onConsumer(uint32_t msgId, const char* roomId, const char* peerId, RTCENGINE_NAMESPACE::MRTCConsumerInfo& consumerInfo);
     void onRender(const char* roomId, const char* peerId,