modify human and tts

Human.py

@@ -1,4 +1,5 @@
 #encoding = utf8
+import copy
 import io
 import logging
 
@@ -25,6 +26,7 @@ from queue import Queue
 
 from tts.EdgeTTS import EdgeTTS
 from tts.TTSBase import TTSBase
+from utils import mirror_index
 
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
@@ -62,16 +64,6 @@ def read_images(img_list):
     return frames
 
 
-def __mirror_index(size, index):
-    # size = len(self.coord_list_cycle)
-    turn = index // size
-    res = index % size
-    if turn % 2 == 0:
-        return res
-    else:
-        return size - res - 1
-
-
 #  python.exe .\inference.py --checkpoint_path .\checkpoints\wav2lip.pth --face
 #  .\face\img00016.jpg --audio .\audio\audio1.wav
 def inference(render_event, batch_size, face_images_path, audio_feat_queue, audio_out_queue, res_frame_queue):
@@ -111,13 +103,13 @@ def inference(render_event, batch_size, face_images_path, audio_feat_queue, audi
         print(f'is_all_silence {is_all_silence}')
         if is_all_silence:
             for i in range(batch_size):
-                res_frame_queue.put((None, __mirror_index(length, index), audio_frames[i*2:i*2+2]))
+                res_frame_queue.put((None, mirror_index(length, index), audio_frames[i*2:i*2+2]))
                 index = index + 1
         else:
             t = time.perf_counter()
             image_batch = []
             for i in range(batch_size):
-                idx = __mirror_index(length, index + i)
+                idx = mirror_index(length, index + i)
                 face = face_list_cycle[idx]
                 image_batch.append(face)
             image_batch, mel_batch = np.asarray(image_batch), np.asarray(mel_batch)
@@ -143,7 +135,7 @@ def inference(render_event, batch_size, face_images_path, audio_feat_queue, audi
                 count_time = 0
 
             for i, res_frame in enumerate(pred):
-                res_frame_queue.put((res_frame, __mirror_index(length, index), audio_frames[i*2 : i*2+2]))
+                res_frame_queue.put((res_frame, mirror_index(length, index), audio_frames[i*2 : i*2+2]))
                 index = index + 1
 
     logging.info('finish inference')
@@ -212,7 +204,7 @@ def datagen(frames, mels):
     # for i, m in enumerate(mels):
     for i in range(mels.qsize()):
         idx = 0 if True else i%len(frames)
-        frame_to_save = frames[__mirror_index(1, i)].copy()
+        frame_to_save = frames[mirror_index(1, i)].copy()
         face, coords = face_det_results[idx].copy()
 
         face = cv2.resize(face, (img_size, img_size))
@@ -294,7 +286,7 @@ def load_audio_from_bytes(byte_data):
 
 class Human:
     def __init__(self):
-        self._fps = 25  # 40 ms per frame
+        self._fps = 50  # 20 ms per frame
         self._batch_size = 16
         self._sample_rate = 16000
         self._stride_left_size = 10
@@ -303,17 +295,27 @@ class Human:
         self._output_queue = mp.Queue()
         self._res_frame_queue = mp.Queue(self._batch_size * 2)
 
-        self._chunk_2_mal = Chunk2Mal(self)
-        self._tts = TTSBase(self)
-        self._infer = Infer(self)
+        full_images, face_frames, coord_frames = self._avatar()
+        self._frame_list_cycle = full_images
+        self._face_list_cycle = face_frames
+        self._coord_list_cycle = coord_frames
+        face_images_length = len(self._face_list_cycle)
+        logging.info(f'face images length: {face_images_length}')
+        print(f'face images length: {face_images_length}')
 
         self.mel_chunks_queue_ = Queue()
         self.audio_chunks_queue_ = Queue()
         self._test_image_queue = Queue()
-        #
-        self._thread = None
-        thread = threading.Thread(target=self.test)
-        thread.start()
+        self._res_render_queue = Queue()
+
+        self._chunk_2_mal = Chunk2Mal(self)
+        self._tts = TTSBase(self)
+        self._infer = Infer(self)
+
+        # #
+        # self._thread = None
+        # thread = threading.Thread(target=self.test)
+        # thread.start()
         # self.test()
         # self.play_pcm()
 
@@ -339,6 +341,21 @@ class Human:
     #     stream.close()
     #     p.terminate()
 
+    def _avatar(self):
+        face_images_path = r'./face/'
+        face_images_path = utils.read_files_path(face_images_path)
+        full_list_cycle = read_images(face_images_path)
+
+        face_det_results = face_detect(full_list_cycle)
+
+        face_frames = []
+        coord_frames = []
+        for face, coord in face_det_results:
+            face_frames.append(face)
+            coord_frames.append(coord)
+
+        return full_list_cycle, face_frames, coord_frames
+
     def inter(self, model, chunks, face_list_cycle, face_det_results, out, j):
         inputs = np.concatenate(chunks)  # [5 * chunk]
         mel = audio.melspectrogram(inputs)
@@ -445,9 +462,11 @@ class Human:
                                                                       'temp/resul_tttt.mp4')
         subprocess.call(command, shell=platform.system() != 'Windows')
 
-
         # gen = datagen(face_list_cycle, self.mel_chunks_queue_)
 
+    def get_face_list_cycle(self):
+        return self._face_list_cycle
+
     def get_fps(self):
         return self._fps
 
@@ -476,12 +495,20 @@ class Human:
             return
         self._tts.push_txt(txt)
 
-    def push_audio_chunk(self, audio_chunk):
-        self._chunk_2_mal.push_chunk(audio_chunk)
+    def put_audio_frame(self, audio_chunk):
+        self._chunk_2_mal.put_audio_frame(audio_chunk)
 
-    def push_mel_chunks_queue(self, mel_chunk):
-        self._infer.push(mel_chunk)
-        # self.audio_chunks_queue_.put(audio_chunk)
+    # def push_audio_chunk(self, audio_chunk):
+    #     self._chunk_2_mal.push_chunk(audio_chunk)
+
+    def push_mel_chunks(self, mel_chunks):
+        self._infer.push(mel_chunks)
+
+    def push_out_put(self, frame, type_):
+        self._infer.push_out_queue(frame, type_)
+
+    def push_mel_chunks_queue(self, audio_chunk):
+        self.audio_chunks_queue_.put(audio_chunk)
 
     def push_feat_queue(self, mel_chunks):
         print("push_feat_queue")
@@ -493,14 +520,38 @@ class Human:
     def push_render_image(self, image):
         self._test_image_queue.put(image)
 
+    def push_res_frame(self, res_frame, idx, audio_frames):
+        self._res_render_queue.put((res_frame, idx, audio_frames))
+
     def render(self):
         try:
             # img, aud = self._res_frame_queue.get(block=True, timeout=.3)
-            img = self._test_image_queue.get(block=True, timeout=.3)
+            # img = self._test_image_queue.get(block=True, timeout=.3)
+            res_frame, idx, audio_frames = self._res_render_queue.get(block=True, timeout=.3)
         except queue.Empty:
             # print('render queue.Empty:')
             return None
-        return img
+
+        if audio_frames[0][1] != 0 and audio_frames[1][1] != 0:
+            combine_frame = self._frame_list_cycle[idx]
+        else:
+            bbox = self._coord_list_cycle[idx]
+            combine_frame = copy.deepcopy(self._frame_list_cycle[idx])
+            y1, y2, x1, x2 = bbox
+            try:
+                res_frame = cv2.resize(res_frame.astype(np.uint8), (x2 - x1, y2 - y1))
+            except:
+                return None
+            # combine_frame = get_image(ori_frame,res_frame,bbox)
+            # t=time.perf_counter()
+            combine_frame[y1:y2, x1:x2] = res_frame
+
+        image = combine_frame
+        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+        return image
+
+
+        # print('blending time:',time.perf_counter()-t)
 
     # def pull_audio_chunk(self):
     #     try:

infer.py

@@ -11,8 +11,8 @@ import torch
 from tqdm import tqdm
 
 import face_detection
-import utils
 from models import Wav2Lip
+from utils import mirror_index
 
 logger = logging.getLogger(__name__)
 
@@ -147,7 +147,8 @@ def datagen_signal(frame, mel, face_det_results):
 class Infer:
     def __init__(self, human):
         self._human = human
-        self._queue = Queue()
+        self._feat_queue = Queue()
+        self._audio_out_queue = Queue()
 
         self._exit_event = Event()
         self._run_thread = Thread(target=self.__on_run)
@@ -155,17 +156,17 @@ class Infer:
         self._run_thread.start()
 
     def __on_run(self):
-        face_images_path = r'./face/'
-        face_images_path = utils.read_files_path(face_images_path)
-        face_list_cycle = read_images(face_images_path)
-        face_images_length = len(face_list_cycle)
-        logging.info(f'face images length: {face_images_length}')
-        print(f'face images length: {face_images_length}')
-
         model = load_model(r'.\checkpoints\wav2lip.pth')
         print("Model loaded")
 
+        face_list_cycle = self._human.get_face_list_cycle()
+
+        # self.__do_run1(face_list_cycle, model)
+        self.__do_run2(face_list_cycle, model)
+
         # frame_h, frame_w = face_list_cycle[0].shape[:-1]
+
+    def __do_run1(self, face_list_cycle, model):
         face_det_results = face_detect(face_list_cycle)
 
         j = 0
@@ -173,7 +174,7 @@ class Infer:
         count = 0
         while self._exit_event.is_set():
             try:
-                m = self._queue.get(block=True, timeout=1)
+                m = self._feat_queue.get(block=True, timeout=1)
             except queue.Empty:
                 continue
 
@@ -202,5 +203,77 @@ class Infer:
                 # out.write(f)
             # print('infer count:', count)
 
-    def push(self, chunk):
-        self._queue.put(chunk)
+    def __do_run2(self, face_list_cycle, model):
+        length = len(face_list_cycle)
+        index = 0
+        count = 0
+        count_time = 0
+        print('start inference')
+        while True:
+            if self._exit_event.is_set():
+                start_time = time.perf_counter()
+                try:
+                    mel_batch = self._feat_queue.get(block=True, timeout=1)
+                except queue.Empty:
+                    continue
+
+                is_all_silence = True
+                audio_frames = []
+                for _ in range(self._human.get_batch_size() * 2):
+                    frame, type_ = self._audio_out_queue.get()
+                    audio_frames.append((frame, type_))
+                    if type_ == 0:
+                        is_all_silence = False
+
+                if is_all_silence:
+                    for i in range(self._human.get_batch_size()):
+                        # res_frame_queue.put((None, mirror_index(length, index), audio_frames[i * 2:i * 2 + 2]))
+                        self._human.push_res_frame(None, mirror_index(length, index), audio_frames[i * 2:i * 2 + 2])
+                        index = index + 1
+                else:
+                    print('infer=======')
+                    t = time.perf_counter()
+                    img_batch = []
+                    for i in range(self._human.get_batch_size()):
+                        idx = mirror_index(length, index + i)
+                        face = face_list_cycle[idx]
+                        img_batch.append(face)
+                    img_batch, mel_batch = np.asarray(img_batch), np.asarray(mel_batch)
+
+                    img_masked = img_batch.copy()
+                    img_masked[:, face.shape[0] // 2:] = 0
+
+                    img_batch = np.concatenate((img_masked, img_batch), axis=3) / 255.
+                    mel_batch = np.reshape(mel_batch, [len(mel_batch), mel_batch.shape[1], mel_batch.shape[2], 1])
+
+                    img_batch = torch.FloatTensor(np.transpose(img_batch, (0, 3, 1, 2))).to(device)
+                    mel_batch = torch.FloatTensor(np.transpose(mel_batch, (0, 3, 1, 2))).to(device)
+
+                    with torch.no_grad():
+                        pred = model(mel_batch, img_batch)
+                    pred = pred.cpu().numpy().transpose(0, 2, 3, 1) * 255.
+
+                    count_time += (time.perf_counter() - t)
+                    count += self._human.batch_size()
+                    # _totalframe += 1
+                    if count >= 100:
+                        print(f"------actual avg infer fps:{count / count_time:.4f}")
+                        count = 0
+                        count_time = 0
+                    for i, res_frame in enumerate(pred):
+                        # self.__pushmedia(res_frame,loop,audio_track,video_track)
+                        # res_frame_queue.put(
+                        #     (res_frame, __mirror_index(length, index), audio_frames[i * 2:i * 2 + 2]))
+                        self._human.push_res_frame(res_frame, mirror_index(length, index),
+                                                   audio_frames[i * 2:i * 2 + 2])
+                        index = index + 1
+                    # print('total batch time:',time.perf_counter()-start_time)
+            else:
+                time.sleep(1)
+        print('musereal inference processor stop')
+
+    def push(self, mel_chunks):
+        self._feat_queue.put(mel_chunks)
+
+    def push_out_queue(self, frame, type_):
+        self._audio_out_queue.put((frame, type_))

tts/Chunk2Mal.py

@@ -1,9 +1,10 @@
 #encoding = utf8
-import ctypes
+
 import logging
 import queue
 import time
 from queue import Queue
+import multiprocessing as mp
 from threading import Thread, Event
 
 import numpy as np
@@ -13,90 +14,69 @@ from audio_render import AudioRender
 
 class Chunk2Mal:
     def __init__(self, human):
-        self._audio_chunk_queue = Queue()
+        # self._audio_chunk_queue = Queue()
         self._human = human
         self._thread = None
 
-        self._chunks = []
-        self._audio_chunks = []
+        self.frames = []
+        self.queue = Queue()
+        # self.output_queue = mp.Queue()
+        # self.feat_queue = mp.Queue(2)
+
         # 320 samples per chunk (20ms * 16000 / 1000)audio_chunk
-        self._chunk_len = self._human.get_audio_sample_rate() // self._human.get_fps()
+        self.chunk = self._human.get_audio_sample_rate() // self._human.get_fps()
 
         self._exit_event = Event()
         self._thread = Thread(target=self._on_run)
         self._exit_event.set()
         self._thread.start()
-        self._audio_render = AudioRender()
+        # self._audio_render = AudioRender()
         self._stream_len = 0
         logging.info('chunk2mal start')
 
-    def _concatenate(self):
-        logging.info('np.concatenate')
-        if len(self._chunks) < 3:
-            logging.info(f'np.concatenate: {len(self._chunks)}')
-            return
-        inputs = np.concatenate(self._chunks)  # [5 * chunk]
-        self._chunks = []
-        mel = audio.melspectrogram(inputs)
-        if np.isnan(mel.reshape(-1)).sum() > 0:
-            raise ValueError(
-                'Mel contains nan! Using a TTS voice? Add a small epsilon noise to the wav file and try again')
-
-        mel_step_size = 16
-        # print('fps:', self._human.get_fps())
-        mel_idx_multiplier = 80. / self._human.get_fps()
-        # print('mel_idx_multiplier:', mel_idx_multiplier)
-        count = 0
-        i = 0
-        while 1:
-            count = count + 1
-            start_idx = int(i * mel_idx_multiplier)
-            print('i', i, 'start_idx', start_idx, 'mel len:', len(mel[0]))
-            if start_idx + mel_step_size > len(mel[0]):
-                self._human.push_mel_chunks_queue(mel[:, len(mel[0]) - mel_step_size:])
-                break
-            self._human.push_mel_chunks_queue(mel[:, start_idx: start_idx + mel_step_size])
-            i += 1
-
-        # wav = np.concatenate(self._audio_chunks)  # [5 * chunk]self._audio_chunks
-        # print('_concatenate', len(wav))
-        # audio.save_chunks([wav], 16000, "./temp/audio/")
-        # wav *= 32767 / max(0.01, np.max(np.abs(wav)))
-        # wav = wav.astype(np.int16)
-        # self._audio_render.write(wav, len(wav))
-        self._audio_chunks = []
-
-        print('mel_chunks count:', count)
-
     def _on_run(self):
         logging.info('chunk2mal run')
         while self._exit_event.is_set():
-            if self._audio_chunk_queue.empty():
-                if len(self._chunks) > 0:
-                    self._concatenate()
-                else:
-                    time.sleep(0.5)
-                continue
-            try:
-                chunk = self._audio_chunk_queue.get(block=True, timeout=1)
-                self._chunks.append(chunk)
-                self._stream_len = self._stream_len + len(chunk)
-                print('Chunk2Mal _stream_len:', self._stream_len)
-                self._audio_chunks.append(chunk.copy())
-
-                # self._human.push_audio_frames(chunk, 0)
-                if len(self._chunks) < 10: # 200ms
-                    continue
-            except queue.Empty:
-                # print('Chunk2Mal queue.Empty')
-                continue
-
-            print('len(self._chunks):', len(self._chunks))
-            self._concatenate()
-
+            self._run_step()
+            time.sleep(0.01)
 
         logging.info('chunk2mal exit')
 
+    def _run_step(self):
+        for _ in range(self._human.get_batch_size() * 2):
+            frame, _type = self.get_audio_frame()
+            self.frames.append(frame)
+            # put to output
+            self._human.push_out_put(frame, _type)
+            # self.output_queue.put((frame, _type))
+        # context not enough, do not run network.
+        if len(self.frames) <= self._human.get_stride_left_size() + self._human.get_stride_right_size():
+            return
+
+        inputs = np.concatenate(self.frames)  # [N * chunk]
+        mel = audio.melspectrogram(inputs)
+        # print(mel.shape[0],mel.shape,len(mel[0]),len(self.frames))
+        # cut off stride
+        left = max(0, self._human.get_stride_left_size() * 80 / 50)
+        right = min(len(mel[0]), len(mel[0]) - self._human.get_stride_right_size() * 80 / 50)
+        mel_idx_multiplier = 80. * 2 / self._human.get_fps()
+        mel_step_size = 16
+        i = 0
+        mel_chunks = []
+        while i < (len(self.frames) - self._human.get_stride_left_size() - self._human.get_stride_right_size()) / 2:
+            start_idx = int(left + i * mel_idx_multiplier)
+            # print(start_idx)
+            if start_idx + mel_step_size > len(mel[0]):
+                mel_chunks.append(mel[:, len(mel[0]) - mel_step_size:])
+            else:
+                mel_chunks.append(mel[:, start_idx: start_idx + mel_step_size])
+            i += 1
+        self._human.push_mel_chunks(mel_chunks)
+        # self.feat_queue.put(mel_chunks)
+
+        # discard the old part to save memory
+        self.frames = self.frames[-(self._human.get_stride_left_size() + self._human.get_stride_right_size()):]
+
     def stop(self):
         if self._exit_event is None:
             return
@@ -106,15 +86,34 @@ class Chunk2Mal:
             self._thread.join()
         logging.info('chunk2mal stop')
 
-    def push_chunk(self, chunk):
-        self._audio_chunk_queue.put(chunk)
+    def pause_talk(self):
+        self.queue.queue.clear()
 
-    def pull_chunk(self):
+    def put_audio_frame(self, audio_chunk): #16khz 20ms pcm
+        self.queue.put(audio_chunk)
+
+    def get_audio_frame(self):
         try:
-            chunk = self._audio_chunk_queue.get(block=True, timeout=1)
-            type = 1
-        except queue.Empty:
-            chunk = np.zeros(self._chunk_len, dtype=np.float32)
+            frame = self.queue.get(block=True, timeout=0.01)
             type = 0
-        return chunk, type
+            # print(f'[INFO] get frame {frame.shape}')
+        except queue.Empty:
+            frame = np.zeros(self.chunk, dtype=np.float32)
+            type = 1
+
+        return frame, type
+
+    def get_audio_out(self):  # get origin audio pcm to nerf
+        return self.output_queue.get()
+
+    def warm_up(self):
+        for _ in range(self._human.get_stride_left_size() + self._human.get_stride_right_size()):
+            audio_frame, _type = self.get_audio_frame()
+            self.frames.append(audio_frame)
+            self.output_queue.put((audio_frame, type))
+        for _ in range(self._human.get_stride_right_size()):
+            self.output_queue.get()
+
+    def get_next_feat(self, block, timeout):
+        return self.feat_queue.get(block, timeout)
 

tts/EdgeTTS.py

@@ -27,6 +27,7 @@ class EdgeTTS(TTSBase):
         stream = self.__create_bytes_stream(self._io_stream)
         stream_len = stream.shape[0]
         index = 0
+
         while stream_len >= self._chunk:
             self._human.push_audio_chunk(stream[index:index + self._chunk])
             stream_len -= self._chunk
@@ -56,4 +57,3 @@ class EdgeTTS(TTSBase):
 
             if chuck['type'] == 'audio':
                 self._io_stream.write(chuck['data'])
-

tts/TTSBase.py

@@ -5,9 +5,7 @@ import time
 
 import edge_tts
 import numpy as np
-import pyaudio
-import soundfile
-import sounddevice
+import soundfile as sf
 import resampy
 import queue
 from io import BytesIO
@@ -16,7 +14,6 @@ from threading import Thread, Event
 
 
 import audio
-from audio_render import AudioRender
 
 logger = logging.getLogger(__name__)
 
@@ -26,17 +23,13 @@ class TTSBase:
         self._human = human
         self._thread = None
         self._queue = Queue()
-        self._io_stream = BytesIO()
-        self._chunk_len = self._human.get_audio_sample_rate() // self._human.get_fps()
+        self.input_stream = BytesIO()
+        self.chunk = self._human.get_audio_sample_rate() // self._human.get_fps()
 
         self._exit_event = Event()
         self._thread = Thread(target=self._on_run)
         self._exit_event.set()
         self._thread.start()
-        # self._pcm_player = pyaudio.PyAudio()
-        # self._pcm_stream = self._pcm_player.open(format=pyaudio.paInt16,
-        #                                          channels=1, rate=24000, output=True)
-        # self._audio_render = AudioRender()
         logging.info('tts start')
 
     def _on_run(self):
@@ -52,94 +45,54 @@ class TTSBase:
     def _request(self, txt):
         voice = 'zh-CN-XiaoyiNeural'
         t = time.time()
-        asyncio.new_event_loop().run_until_complete(self.__on_request(voice, txt))
-        logger.info(f'edge tts time:{time.time() - t : 0.4f}s')
+        asyncio.new_event_loop().run_until_complete(self.__main(voice, txt))
+        print(f'-------edge tts time:{time.time() - t:.4f}s')
 
-        self._io_stream.seek(0)
-        stream = self.__create_bytes_stream(self._io_stream)
-        audio.save_chunks([stream], 16000, './temp/audio/')
-        # wav = audio.split_audio(stream, 16000, 0.04)
+        self.input_stream.seek(0)
+        stream = self.__create_bytes_stream(self.input_stream)
+        streamlen = stream.shape[0]
+        idx = 0
+        print('-------tts start push chunk')
+        while streamlen >= self.chunk:
+            self._human.put_audio_frame(stream[idx:idx + self.chunk])
+            streamlen -= self.chunk
+            idx += self.chunk
+        # if streamlen>0:  #skip last frame(not 20ms)
+        #    self.queue.put(stream[idx:])
+        self.input_stream.seek(0)
+        self.input_stream.truncate()
+        print('-------tts finish push chunk')
 
-        # audio.save_chunks(wav, 16000, './temp/audio/')
-        # audio.save_wav(stream, "./temp/audio/test1.wav", 16000)
-        stream_len = stream.shape[0]
-
-        # wav = stream #np.concatenate(stream)  # [5 * chunk]self._audio_chunks
-        # print('_concatenate', len(wav))
-
-        # self._audio_chunks = []
-
-        print("stream_len:", stream_len, " _chunk_len:", self._chunk_len)
-        index = 0
-        segment = 0
-        while stream_len >= self._chunk_len:
-            audio_chunk = stream[index:index + self._chunk_len]
-            self._human.push_audio_chunk(audio_chunk)
-            stream_len -= self._chunk_len
-            index += self._chunk_len
-            segment = segment + 1
-
-        if stream_len > 0:
-            audio_chunk = stream[index:index + stream_len]
-            self._human.push_audio_chunk(audio_chunk)
-            segment = segment + 1
-
-        print("segment:", segment)
-        self._io_stream.seek(0)
-        self._io_stream.truncate()
-
-    def __create_bytes_stream(self, io_stream):
-        stream, sample_rate = soundfile.read(io_stream)
-        logger.info(f'tts audio stream {sample_rate} : {stream.shape}')
+    def __create_bytes_stream(self, byte_stream):
+            # byte_stream=BytesIO(buffer)
+            stream, sample_rate = sf.read(byte_stream)  # [T*sample_rate,] float64
+            print(f'[INFO]tts audio stream {sample_rate}: {stream.shape}')
             stream = stream.astype(np.float32)
 
             if stream.ndim > 1:
-            logger.warning(f'tts audio has {stream.shape[1]} channels, only use the first')
-            stream = stream[:, 1]
+                print(f'[WARN] audio has {stream.shape[1]} channels, only use the first.')
+                stream = stream[:, 0]
 
             if sample_rate != self._human.get_audio_sample_rate() and stream.shape[0] > 0:
-            logger.warning(f'tts audio sample rate is {sample_rate}, resample to {self._human.get_audio_sample_rate() }')
+                print(f'[WARN] audio sample rate is {sample_rate}, resampling into {self._human.get_audio_sample_rate()}.')
                 stream = resampy.resample(x=stream, sr_orig=sample_rate, sr_new=self._human.get_audio_sample_rate())
 
             return stream
 
-    async def __on_request(self, voice, txt):
-        communicate = edge_tts.Communicate(txt, voice)
+    async def __main(self, voicename: str, text: str):
+        communicate = edge_tts.Communicate(text, voicename)
+
+        #with open(OUTPUT_FILE, "wb") as file:
         first = True
-        total_data = b''
-        CHUNK_SIZE = self._chunk_len
         async for chunk in communicate.stream():
-            if chunk["type"] == "audio" and chunk["data"]:
-                data = chunk['data']
-                self._io_stream.write(data)
+            if first:
+                first = False
+            if chunk["type"] == "audio":
+                #self.push_audio(chunk["data"])
+                self.input_stream.write(chunk["data"])
+                #file.write(chunk["data"])
             elif chunk["type"] == "WordBoundary":
                 pass
-                '''
-                total_data += chunk["data"]
-                if len(total_data) >= CHUNK_SIZE:
-                #     print(f"Time elapsed: {time.time() - start_time:.2f} seconds")  # Print time
-                    audio_data = AudioSegment.from_mp3(BytesIO(total_data[:CHUNK_SIZE])) #.raw_data
-                    audio_data = audio_data.set_frame_rate(self._human.get_audio_sample_rate())
-                    # self._human.push_audio_chunk(audio_data)
-                    self._pcm_stream.write(audio_data.raw_data)
-                    # play_audio(total_data[:CHUNK_SIZE], stream)  # Play first CHUNK_SIZE bytes
-                    total_data = total_data[CHUNK_SIZE:]  # Remove played data
-                '''
-
-            # if first:
-            #     first = False
-
-            # if chuck['type'] == 'audio':
-            #     # self._io_stream.write(chuck['data'])
-            #     self._io_stream.write(AudioSegment.from_mp3(BytesIO(total_data[:CHUNK_SIZE])).raw_data)
-
-        # if len(total_data) > 0:
-             # self._pcm_stream.write(AudioSegment.from_mp3(BytesIO(total_data)).raw_data)
-             # audio_data = AudioSegment.from_mp3(BytesIO(total_data))  # .raw_data
-             # audio_data = audio_data.set_frame_rate(self._human.get_audio_sample_rate())
-             # self._pcm_stream.write(audio_data.raw_data)
-             # self._human.push_audio_chunk(audio_data)
-        # self._io_stream.write(AudioSegment.from_mp3(BytesIO(total_data)).raw_data)
 
     def stop(self):
         self._pcm_stream.stop_stream()

ui.py

@@ -44,7 +44,7 @@ class App(customtkinter.CTk):
         # self.logo_label.grid(row=0, column=0, padx=20, pady=(20, 10))
 
         self.entry = customtkinter.CTkEntry(self, placeholder_text="输入内容")
-        self.entry.insert(0, "基本信息,北京九零科技有限公司，成立于2015年，位于北京市，是一家以从事科技推广和应用服务业为主的企业。企业注册资本500万人民币。")
+        self.entry.insert(0, "你好，我是中国湘西人。")
         self.entry.grid(row=2, column=0, columnspan=2, padx=(20, 0), pady=(20, 20), sticky="nsew")
 
         self.main_button_1 = customtkinter.CTkButton(master=self, fg_color="transparent", border_width=2,
@@ -83,7 +83,7 @@ class App(customtkinter.CTk):
             self.after(100, self._render)
             return
 
-        self.play_audio()
+        # self.play_audio()
         iheight, iwidth = image.shape[0], image.shape[1]
         width = self.winfo_width()
         height = self.winfo_height()

utils.py

@@ -10,3 +10,13 @@ def read_files_path(path):
             file_paths.append(path + file)
     return file_paths
 
+
+def mirror_index(size, index):
+    # size = len(self.coord_list_cycle)
+    turn = index // size
+    res = index % size
+    if turn % 2 == 0:
+        return res
+    else:
+        return size - res - 1
+