Update app.py

app.py

@@ -1,3 +1,4 @@
+import numpy as np
 import logging
 logging.getLogger('numba').setLevel(logging.WARNING)
 import IPython.display as ipd
@@ -71,30 +72,59 @@ def selection(speaker):
     elif speaker == "派蒙":
         spk = 16
         return spk
-    
+def is_japanese(string):
+        for ch in string:
+            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
+                return True
+        return False        
 def sle(language,tts_input0):
     if language == "中文":
         tts_input1 = "[ZH]" + tts_input0.replace('\n','。').replace(' ',',') + "[ZH]"
         return tts_input1
-    if language == "英文":
-        tts_input1 = "[EN]" + tts_input0.replace('\n','.').replace(' ',',') + "[EN]"
+    if language == "自动":
+        tts_input1 = f"[JA]{tts_input0}[JA]" if is_japanese(tts_input0) else f"[ZH]{tts_input0}[ZH]"
         return tts_input1
     elif language == "日文":
         tts_input1 = "[JA]" + tts_input0.replace('\n','。').replace(' ',',') + "[JA]"
         return tts_input1
+def extrac(text):
+    result_list = re.split(r'\n', text)
+    final_list = []
+    for i in result_list:
+        i = i.replace('\n','').replace(' ','')
+        if len(i)>0:
+            if len(i) > 20:
+                try:
+                    cur_list = re.split(r'。', i)
+                    for i in cur_list:
+                        if len(i)>0:
+                            final_list.append(i+'。')
+                except:
+                    pass
+            final_list.append(i)
+    final_list = [x for x in final_list if x != '']
+    print(final_list)
+    return final_list
 def infer(language,text,speaker_id, n_scale= 0.667,n_scale_w = 0.8, l_scale = 1 ):
     speaker_id = int(selection(speaker_id))
-    stn_tst = get_text(sle(language,text), hps_ms)
-    with torch.no_grad():
-        x_tst = stn_tst.unsqueeze(0).to(dev)
-        x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(dev)
-        sid = torch.LongTensor([speaker_id]).to(dev)
-        t1 = time.time()
-        audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=n_scale, noise_scale_w=n_scale_w, length_scale=l_scale)[0][0,0].data.cpu().float().numpy()
-        t2 = time.time()
-        spending_time = "推理时间："+str(t2-t1)+"s" 
-        print(spending_time)
-    return (hps_ms.data.sampling_rate, audio) 
+    final_list = extrac(text)
+    audio_fin = []
+    c = 0
+    for sentence in final_list:
+        c +=1
+        try:
+            with torch.no_grad():
+                x_tst = stn_tst.unsqueeze(0).to(dev)
+                x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(dev)
+                sid = torch.LongTensor([speaker_id]).to(dev)
+                t1 = time.time()
+                audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=n_scale, noise_scale_w=n_scale_w, length_scale=l_scale)[0][0,0].data.cpu().float().numpy()
+                t2 = time.time()
+                spending_time = "第"+str(c)+"句的推理时间为："+str(t2-t1)+"s" 
+                print(spending_time)
+        except:
+            print('存在非法字符')
+    return (hps_ms.data.sampling_rate, np.concatenate(audio_fin)) 
 lan = ["中文","日文","英文"]
 idols = ["高咲侑","歩夢","かすみ","しずく","果林","愛","せつ菜","璃奈","栞子","エマ","ランジュ","ミア","派蒙"]
 dev = torch.device("cpu")
@@ -106,17 +136,17 @@ net_g_ms = ONNXVITS_infer.SynthesizerTrn(
     n_speakers=hps_ms.data.n_speakers,
     **hps_ms.model)
 _ = net_g_ms.eval()
-_ = utils.load_checkpoint("lovelive/G_330000.pth", net_g_ms)
+_ = utils.load_checkpoint("lovelive/G_525000.pth", net_g_ms)
 app = gr.Blocks()
 with app:
     with gr.Tabs():
 
-        with gr.TabItem("New model"):
+        with gr.TabItem("虹团vits模型，现可按句分割实现长文本合成"):
 
-            tts_input1 = gr.TextArea(label="新模型，老版本在lovelive文件夹中", value="为什么你会这么熟练啊！你和雪菜亲过多少次了啊！？你到底要把我甩开多远你才甘心啊！？")
-            language = gr.Dropdown(label="选择语言",choices=lan, value="中文", interactive=True)
+            tts_input1 = gr.TextArea(label="去标贝新模型，老版本在lovelive文件夹中", value="你一点也不在意我吗？一点也不会吗？完全不会？我对你来说不重要吗？只是朋友吗？普通的朋友吗？我希望自己不是普通的朋友，就算比普通好一点也好，我想成为不普通的……朋友……嗳，岛村，我该怎么做才好？嗳。岛村，你有在听吗？求你听我说话。你听到我的声音有什么想法吗？会有吗？你要说会感到放心，还是什么都好，拜托有点想法。我希望你可以有点想法，还是说我不该期待这种事？岛村！我就是要岛村啊，我啊，就是想要跟岛村待在一起。我不需要岛村以外的人，不需要……只要有岛村就好。我没有很任性喔，我只说比普通好一点，好一点而已啊。其他人根本就不重要，也不需要，我希望那些人都���以滚远一点，可是你为什么要去他们那边呢？求你来我这边，来我这边，待在我身边，不要离开我。不行，在岛村身旁的只能是我，我希望是我，我想待在你身边，拜托你让我待在你身边……那个女生是谁？我不认识她啊。我不想看到你变成我不认识的岛村，我想了解岛村的一切，也讨厌有我不想知道的事情存在，可是我更讨厌自己不知道，会更难受。会很难受，很痛苦，很痛苦……岛村……我想问你要不要一起出去玩，也想去祭典啊，我很想去啊，可是岛村为什么会跟那个女生一起去？为什么跟她一起出去玩？岛村你现在在哪里？有跟谁在一起吗？岛村，岛村……嗳，你有在听吗？")
+            language = gr.Dropdown(label="选择语言，目前勉强可以做到自动识别",choices=lan, value="自动", interactive=True)
             para_input1 = gr.Slider(minimum= 0.01,maximum=1.0,label="更改噪声比例", value=0.667)
-            para_input2 = gr.Slider(minimum= 0.01,maximum=1.0,label="更改噪声偏差", value=0.8)
+            para_input2 = gr.Slider(minimum= 0.01,maximum=1.0,label="更改噪声偏差，调小可以缩短间隔", value=0.8)
             para_input3 = gr.Slider(minimum= 0.1,maximum=10,label="更改时间比例", value=1)
             tts_submit = gr.Button("Generate", variant="primary")
             speaker1 = gr.Dropdown(label="选择说话人",choices=idols, value="歩夢", interactive=True)