modify human

Human.py

@@ -81,7 +81,7 @@ def inference(render_event, batch_size, face_images_path, audio_feat_queue, audi
     logging.info('start inference')
     print(f'start inference: {render_event.is_set()}')
     while render_event.is_set():
-        print('start inference')
+        mel_batch = []
         try:
             mel_batch = audio_feat_queue.get(block=True, timeout=1)
         except queue.Empty:
@@ -139,6 +139,7 @@ def inference(render_event, batch_size, face_images_path, audio_feat_queue, audi
 
 class Human:
     def __init__(self):
+        self._text = None
         self._tts = None
         self._fps = 50  # 20 ms per frame
         self._batch_size = 16
@@ -201,18 +202,18 @@ class Human:
         self._chunk_2_mal.push_chunk(chunk)
 
     def push_feat_queue(self, mel_chunks):
-        print("21")
+        print("push_feat_queue")
         self._feat_queue.put(mel_chunks)
-        print("22")
 
     def push_audio_frames(self, chunk, type_):
+        print("push_audio_frames")
         self._output_queue.put((chunk, type_))
 
     def render(self):
         try:
             img, aud = self._res_frame_queue.get(block=True, timeout=.3)
         except queue.Empty:
-            print('queue.Empty:')
+            # print('render queue.Empty:')
             return None
         return img
 

tts/Chunk2Mal.py

@@ -23,14 +23,16 @@ class Chunk2Mal:
             try:
                 chunk, type_ = self.pull_chunk()
                 self._chunks.append(chunk)
-                self._human.push_audio_frames((chunk, type_))
-                print("1")
+                self._human.push_audio_frames(chunk, type_)
             except queue.Empty:
+                # print('Chunk2Mal queue.Empty')
                 continue
 
             if len(self._chunks) <= self._human.get_stride_left_size() + self._human.get_stride_right_size():
+                # print('Chunk2Mal queue.Empty')
                 continue
 
+            logging.info('np.concatenate')
             inputs = np.concatenate(self._chunks)  # [N * chunk]
             mel = audio.melspectrogram(inputs)
             left = max(0, self._human.get_stride_left_size() * 80 / 50)
@@ -41,7 +43,6 @@ class Chunk2Mal:
             mel_chunks = []
             while i < (len(self._chunks) - self._human.get_stride_left_size()
                        - self._human.get_stride_right_size()) / 2:
-                print("14")
                 start_idx = int(left + i * mel_idx_multiplier)
                 # print(start_idx)
                 if start_idx + mel_step_size > len(mel[0]):
@@ -49,13 +50,10 @@ class Chunk2Mal:
                 else:
                     mel_chunks.append(mel[:, start_idx: start_idx + mel_step_size])
                 i += 1
-                print("13")
             self._human.push_feat_queue(mel_chunks)
-            print("15")
 
             # discard the old part to save memory
             self._chunks = self._chunks[-(self._human.get_stride_left_size() + self._human.get_stride_right_size()):]
-            print("12")
 
         logging.info('chunk2mal exit')