[ADD]add logic of loop frame

human/audio_inference_handler.py

@@ -21,14 +21,16 @@ current_file_path = os.path.dirname(os.path.abspath(__file__))
 
 
 class AudioInferenceHandler(AudioHandler):
-    def __init__(self, context, handler):
+    def __init__(self, context, handler, person_config):
         super().__init__(context, handler)
 
         EventBus().register('stop', self._on_stop)
         self._mal_queue = SyncQueue(1, "AudioInferenceHandler_Mel")
         self._audio_queue = SyncQueue(context.batch_size * 2, "AudioInferenceHandler_Audio")
+        self.person_config = person_config
 
         self._is_running = True
+        self.last_direction = 1
         self._exit_event = Event()
         self._run_thread = Thread(target=self.__on_run, name="AudioInferenceHandlerThread")
         self._exit_event.set()
@@ -102,12 +104,12 @@ class AudioInferenceHandler(AudioHandler):
                     break
 
                 if is_all_silence:
-                    for i in range(batch_size):
+                    frame_indexes, self.last_direction = human_status.get_index_v2(self.person_config["frame_config"], self.last_direction, batch_size) # [1,3,4]
+                    for i, frame_idx in zip(range(batch_size), frame_indexes):
                         if not self._is_running:
                             break
                         # self.on_next_handle((None, mirror_index(length, index),
-                        self.on_next_handle((None, human_status.get_index(),
+                        self.on_next_handle((None,frame_idx, audio_frames[i * 2:i * 2 + 2]), 0)
-                                             audio_frames[i * 2:i * 2 + 2]), 0)
                         # index = index + 1
                 else:
                     human_status.start_talking()
@@ -117,12 +119,12 @@ class AudioInferenceHandler(AudioHandler):
                     img_batch = []
                     index_list = []
                     # for i in range(batch_size):
+                    frame_indexes,self.last_direction = human_status.get_index_v2(self.person_config["frame_config"], self.last_direction, batch_size) # [1,3,4]
                     # TODO: 推理状态下获取循环帧逻辑
-                    for i in range(len(mel_batch)):
+                    for i, frame_idx in zip(range(len(mel_batch)), frame_indexes):
                         # idx = mirror_index(length, index + i)
-                        idx = human_status.get_index()
+                        index_list.append(frame_idx)
-                        index_list.append(idx)
+                        face = face_list_cycle[frame_idx]
-                        face = face_list_cycle[idx]
                         img_batch.append(face)
 
                     # print('orign img_batch:', len(img_batch), 'origin mel_batch:', len(mel_batch))

human/huaman_status.py

@@ -52,11 +52,10 @@ class HumanStatus:
         self._current_frame = (self._current_frame + 1) % self._total_frames
         return index
     
-    def get_index_v2(self):
+    def get_index_v2(self, frame_config:list, last_direction:int=1, batch_size:int=5):
         """
         """
-        frame_config = []
+        audio_frame_length = batch_size
-        audio_frame_length = batch_size = 5
         is_silent = True if not self._is_talking else False
         first_speak = True if self._is_talking and self.last_frame_talking_status == "silent" else False
         last_speak = True if self.last_frame_talking_status == "talk" and self.next_frame_talking_status == "silent" else False
@@ -71,7 +70,7 @@ class HumanStatus:
         )
         startfrom = start_idx_list[-1]
         # 一次返回一个batch帧号
-        return start_idx_list
+        return start_idx_list, last_direction
 
 
     def start_talking(self):

human/human_context.py

@@ -47,6 +47,10 @@ class HumanContext:
         self._m_frames = m_frames
         self._inv_m_frames = inv_m_frames
         face_images_length = len(self._face_list_cycle)
+        # TODO: get person config
+        self.person_config ={
+            "frame_config": [[1,face_frames-1, True]],
+        }
         logging.info(f'face images length: {face_images_length}')
         print(f'face images length: {face_images_length}')
 
@@ -113,7 +117,7 @@ class HumanContext:
 
     def build(self, render_handler):
         self._render_handler = render_handler
-        self._infer_handler = AudioInferenceHandler(self, self._render_handler)
+        self._infer_handler = AudioInferenceHandler(self, self._render_handler, self.person_config)
         self._mal_handler = AudioMalHandler(self,  self._infer_handler)
         self._tts_handle = TTSAudioSplitHandle(self,  self._mal_handler)
         self._tts = TTSEdgeHttp(self._tts_handle)