modify ipc

human/__init__.py

@@ -4,4 +4,5 @@ from .human_context import HumanContext
 from .audio_mal_handler import AudioMalHandler
 from .audio_inference_handler import AudioInferenceHandler
 from .audio_inference_onnx_handler import AudioInferenceOnnxHandler
+from .huaman_status import HumanStatusEnum, HumanStatus
 from .human_render import HumanRender

human/audio_inference_handler.py

@@ -6,10 +6,12 @@ import time
 from queue import Queue
 from threading import Event, Thread
 
+import cv2
 import numpy as np
 import torch
 
 from eventbus import EventBus
+from human import HumanStatus
 from human_handler import AudioHandler
 from utils import load_model, mirror_index, get_device, SyncQueue
 
@@ -67,12 +69,13 @@ class AudioInferenceHandler(AudioHandler):
         logger.info("Model loaded")
 
         face_list_cycle = self._context.face_list_cycle
-
         length = len(face_list_cycle)
         index = 0
         count = 0
         count_time = 0
         logger.info('start inference')
+        silence_length = 133
+        human_status = HumanStatus(length, silence_length)
 
         device = get_device()
         logger.info(f'use device:{device}')
@@ -107,11 +110,13 @@ class AudioInferenceHandler(AudioHandler):
                     for i in range(batch_size):
                         if not self._is_running:
                             break
-                        self.on_next_handle((None, mirror_index(length, index), audio_frames[i * 2:i * 2 + 2]),
+                        # self.on_next_handle((None, mirror_index(silence_length, index),
-                                            0)
+                        self.on_next_handle((None, human_status.get_index(),
+                                             audio_frames[i * 2:i * 2 + 2]), 0)
                         index = index + 1
                 else:
                     logger.info(f'infer======= {current_text}')
+                    human_status.try_to_talk()
                     t = time.perf_counter()
                     img_batch = []
                     # for i in range(batch_size):

human/audio_inference_onnx_handler.py

@@ -5,6 +5,8 @@ import queue
 import time
 from threading import Event, Thread
 
+import cv2
+
 # from gfpgan import GFPGANer
 from eventbus import EventBus
 from human_handler import AudioHandler
@@ -93,9 +95,9 @@ class AudioInferenceOnnxHandler(AudioHandler):
         gfpgan_model = load_gfpgan_model(gfpgan_model_path)
 
         face_list_cycle = self._context.face_list_cycle
-
-
         length = len(face_list_cycle)
+        for i in range(length):
+            cv2.imwrite(f'face_{i}.png', face_list_cycle[i])
         index = 0
         count = 0
         count_time = 0
@@ -157,18 +159,6 @@ class AudioInferenceOnnxHandler(AudioHandler):
                     onnx_out = model_g.run(onnx_names, onnx_input)[0]
                     pred = onnx_out
 
-                    # onnxruntime_inputs = {"audio_seqs__0": mel_batch, }
-                    # onnxruntime_names = [output.name for output in model_a.get_outputs()]
-                    # embeddings = model_a.run(onnxruntime_names, onnxruntime_inputs)[0]
-                    #
-                    # onnxruntime_inputs = {"audio_embedings__0": embeddings, "img_seqs__1": img_batch}
-                    # onnxruntime_names = [output.name for output in model_g.get_outputs()]
-                    #
-                    # start_model = time.time()
-                    # onnxruntime_output = model_g.run(onnxruntime_names, onnxruntime_inputs)[0]
-                    # end_model = time.time()
-                    # pred = onnxruntime_output
-
                     count_time += (time.perf_counter() - t)
                     count += batch_size
 

human/huaman_status.py

@@ -0,0 +1,36 @@
+#encoding = utf8
+
+import logging
+
+
+from enum import Enum
+
+
+class HumanStatusEnum(Enum):
+    silence = 1
+    talking = 2
+
+
+class HumanStatus:
+    def __init__(self, total_frames=0, last_silence_frame=0):
+        self._status = HumanStatusEnum.silence
+        self._total_frames = total_frames
+        self._last_silence_frame = last_silence_frame
+        self._current_frame = 0
+
+    def get_status(self):
+        return self._status
+
+    def set_status(self, status):
+        self._status = status
+        return self._status
+
+    def try_to_talk(self):
+        if self._status == HumanStatusEnum.silence:
+            if self._current_frame - self._last_silence_frame < 0:
+                return False
+            self._status = HumanStatusEnum.talking
+        return True
+
+    def get_index(self):
+        return self._total_frames

ipc/ipc_mem.py

@@ -0,0 +1,62 @@
+#encoding = utf8
+
+import logging
+import os
+
+from ctypes import *
+
+current = os.path.dirname(__file__)
+dynamic_path = os.path.join(current, 'ipc.dll')
+
+
+class IPCMem:
+    def __init__(self, sender, receiver):
+        self.__ipc_obj = WinDLL(dynamic_path)
+        print(self.__ipc_obj)
+        if self.__ipc_obj is not None:
+            self.__ipc_obj.initialize.argtypes = [c_char_p, c_char_p]
+            self.__ipc_obj.initialize.restype = c_bool
+            print('IPCUtil init', sender.encode('utf-8'), receiver.encode('utf-8'))
+            self.__init = self.__ipc_obj.initialize(sender.encode('utf-8'), receiver.encode('utf-8'))
+            print('IPCUtil init', self.__init)
+
+    def __del__(self):
+        print('IPCUtil __del__')
+        if self.__ipc_obj is None:
+            return
+        if self.__init:
+            self.__ipc_obj.uninitialize()
+
+    def listen(self):
+        if not self.__init:
+            return False
+        self.__ipc_obj.listen.restype = c_bool
+        return self.__ipc_obj.listen()
+
+    def send_text(self, data):
+        if not self.__init:
+            return False
+        self.__ipc_obj.send.argtypes = [c_char_p, c_uint]
+        self.__ipc_obj.send.restype = c_bool
+        send_data = data.encode('utf-8')
+        send_len = len(send_data) + 1
+        if not self.__ipc_obj.send(send_data, send_len):
+            self.__ipc_obj.reConnect()
+        return True
+
+    def send_binary(self, data, size):
+        if not self.__init:
+            return False
+        self.__ipc_obj.send.argtypes = [c_char_p, c_uint]
+        self.__ipc_obj.send.restype = c_bool
+        data_ptr = cast(data, c_char_p)
+        return self.__ipc_obj.send(data_ptr, size)
+
+    def set_reader_callback(self, callback):
+        if not self.__init:
+            return False
+        CALLBACK_TYPE = CFUNCTYPE(None, c_char_p, c_uint)
+        self.c_callback = CALLBACK_TYPE(callback)  # Store the callback to prevent garbage collection
+        self.__ipc_obj.setReaderCallback.argtypes = [CALLBACK_TYPE]
+        self.__ipc_obj.setReaderCallback.restype = c_bool
+        return self.__ipc_obj.setReaderCallback(self.c_callback)

ipc/ipc_util.py

@@ -38,7 +38,7 @@ class IPCUtil:
         self.__ipc_obj.send.restype = c_bool
         send_data = data.encode('utf-8')
         send_len = len(send_data) + 1
-        if not self.__ipc_obj.send(send_data, send_len):
+        if not self.__ipc_obj.trySend(send_data, send_len):
             self.__ipc_obj.reConnect()
         return True
 
@@ -48,7 +48,7 @@ class IPCUtil:
         self.__ipc_obj.send.argtypes = [c_char_p, c_uint]
         self.__ipc_obj.send.restype = c_bool
         data_ptr = cast(data, c_char_p)
-        return self.__ipc_obj.send(data_ptr, size)
+        return self.__ipc_obj.trySend(data_ptr, size)
 
     def set_reader_callback(self, callback):
         if not self.__init:

main.py

@@ -15,4 +15,5 @@ if __name__ == '__main__':
     logger.info('------------start------------')
     render = IpcRender()
     render.run()
+    render.stop()
     logger.info('------------finish------------')

ui/ipc_render.py

@@ -64,7 +64,8 @@ class IpcRender:
         logger.info('ipc render exit')
 
     def stop(self):
-        pass
+        if self._human_context is not None:
+            self._human_context.stop()
 
     def on_render(self, image):
         self._queue.put(image)