Update app.py

app.py

@@ -1,17 +1,12 @@
 import gradio as gr
-import os
-import random
-import IPython.display as ipd
+import torch
 import commons
 import utils
-import json
-import torch
-import tempfile
-import numpy as np
 from models import SynthesizerTrn
 from text.symbols import symbols
 from text import text_to_sequence
-from scipy.io.wavfile import write
+import numpy as np
+
 
 def get_text(text, hps):
     text_norm = text_to_sequence(text, hps.data.text_cleaners)
@@ -19,16 +14,6 @@ def get_text(text, hps):
         text_norm = commons.intersperse(text_norm, 0)
     text_norm = torch.LongTensor(text_norm)
     return text_norm
-
-def get_text_byroma(text, hps):
-    text_norm = []
-    for i in text:
-        text_norm.append(symbols.index(i))
-    if hps.data.add_blank:
-        text_norm = commons.intersperse(text_norm, 0)
-    text_norm = torch.LongTensor(text_norm)
-    return text_norm
-
 hps = utils.get_hparams_from_file("./configs/leo.json")
 net_g = SynthesizerTrn(
     len(symbols),
@@ -37,61 +22,90 @@ net_g = SynthesizerTrn(
     n_speakers=hps.data.n_speakers,
     **hps.model)
 _ = net_g.eval()
-_ = utils.load_checkpoint("logs/leo/G_4000.pth", net_g, None)
 
-# 随机抽取情感参考音频的根目录
-random_emotion_root = "wavs"
-emotion_dict = json.load(open("configs/leo.json", "r"))
-
-def tts(txt, emotion, temp_file_path):
-    """emotion为参考情感音频路径或random_sample（随机抽取）"""
-    if roma:
-        stn_tst = get_text_byroma(txt, hps)
-    else:
-        stn_tst = get_text(txt, hps)
+_ = utils.load_checkpoint("./logs/leo/G_4000.pth", net_g, None)
+all_emotions = np.load("all_emotions.npy")
+emotion_dict = {
+    "小声": 0,
+    "激动": 1,
+    "平静1": 2,
+    "平静2": 3
+}
+import random
+def tts(txt, emotion):
+    stn_tst = get_text(txt, hps)
+    randsample = None
     with torch.no_grad():
         x_tst = stn_tst.unsqueeze(0)
         x_tst_lengths = torch.LongTensor([stn_tst.size(0)])
         sid = torch.LongTensor([0])
-        if os.path.exists(f"{emotion}.emo.npy"):
-            emo = torch.FloatTensor(np.load(f"{emotion}.emo.npy")).unsqueeze(0)
+        if type(emotion) ==int:
+            emo = torch.FloatTensor(all_emotions[emotion]).unsqueeze(0)
+        elif emotion == "random":
+            emo = torch.randn([1,1024])
         elif emotion == "random_sample":
-            while True:
-                rand_wav = random.sample(os.listdir(random_emotion_root), 1)[0]
-                if rand_wav.endswith('wav') and os.path.exists(f"{random_emotion_root}/{rand_wav}.emo.npy"):
-                    break
-            emo = torch.FloatTensor(np.load(f"{random_emotion_root}/{rand_wav}.emo.npy")).unsqueeze(0)
-            print(f"{random_emotion_root}/{rand_wav}")
+            randint = random.randint(0, all_emotions.shape[0])
+            emo = torch.FloatTensor(all_emotions[randint]).unsqueeze(0)
+            randsample = randint
         elif emotion.endswith("wav"):
             import emotion_extract
             emo = torch.FloatTensor(emotion_extract.extract_wav(emotion))
         else:
-            print("emotion参数不正确")
+            emo = torch.FloatTensor(all_emotions[emotion_dict[emotion]]).unsqueeze(0)
+
+        audio = net_g.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=0.667, noise_scale_w=0.8, length_scale=1, emo=emo)[0][0,0].data.float().numpy()
+    return audio, randsample
+
 
-        audio = net_g.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=0.667, noise_scale_w=0.8, length_scale=1.2, emo=emo)[0][0,0].data.float().numpy()
+def tts1(text, emotion):
+    if len(text) > 150:
+        return "Error: Text is too long", None
+    audio, _ = tts(text, emotion)
+    return "Success", (hps.data.sampling_rate, audio)
 
-        # Save the numpy array as a temporary file
-        write(temp_file_path, hps.data.sampling_rate, audio)
+def tts2(text):
+    if len(text) > 150:
+        return "Error: Text is too long", None
+    audio, randsample = tts(text, "random_sample")
 
-        # Display the audio
-        ipd.display(ipd.Audio(temp_file_path, rate=hps.data.sampling_rate, normalize=False))
+    return str(randsample), (hps.data.sampling_rate, audio)
 
-        # Delete the temporary file
-        os.remove(temp_file_path)
+def tts3(text, sample):
+    if len(text) > 150:
+        return "Error: Text is too long", None
+    try:
+        audio, _ = tts(text, int(sample))
+        return "Success", (hps.data.sampling_rate, audio)
+    except:
+        return "输入参数不为整数或其他错误", None
+app = gr.Blocks()
+with app:
+    with gr.Tabs():
+        with gr.TabItem("使用预制情感合成"):
+            tts_input1 = gr.TextArea(label="日语文本", value="こんにちは。私わあやちねねです。")
+            tts_input2 = gr.Dropdown(label="情感", choices=list(emotion_dict.keys()),  value="平静1")
+            tts_submit = gr.Button("合成音频", variant="primary")
+            tts_output1 = gr.Textbox(label="Message")
+            tts_output2 = gr.Audio(label="Output")
+            tts_submit.click(tts1, [tts_input1, tts_input2], [tts_output1, tts_output2])
+        with gr.TabItem("随机抽取训练集样本作为情感参数"):
+            tts_input1 = gr.TextArea(label="日语文本", value="こんにちは。私わあやちねねです。")
+            tts_submit = gr.Button("合成音频", variant="primary")
+            tts_output1 = gr.Textbox(label="随机样本id（可用于第三个tab中合成）")
+            tts_output2 = gr.Audio(label="Output")
+            tts_submit.click(tts2, [tts_input1], [tts_output1, tts_output2])
 
-    return audio
+        with gr.TabItem("使用情感样本id作为情感参数"):
 
-def generate_audio(txt, emotion):
-    # Create a temporary file
-    temp_file = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
-    temp_file_path = temp_file.name
-    audio = tts(txt, emotion, temp_file_path)
-    return audio
+            tts_input1 = gr.TextArea(label="日语文本", value="こんにちは。私わあやちねねです。")
+            tts_input2 = gr.Number(label="情感样本id", value=2004)
+            tts_submit = gr.Button("合成音频", variant="primary")
+            tts_output1 = gr.Textbox(label="Message")
+            tts_output2 = gr.Audio(label="Output")
+            tts_submit.click(tts3, [tts_input1, tts_input2], [tts_output1, tts_output2])
 
-input_text = gr.inputs.Textbox(label="输入文本")
-input_emotion = gr.inputs.Dropdown(choices=["random_sample"] + os.listdir(random_emotion_root), label="参考情感音频")
-output_audio = gr.outputs.Audio(type="numpy", label="合成音频")
+        with gr.TabItem("使用参考音频作为情感参数"):
+            tts_input1 = gr.TextArea(label="text", value="暂未实现")
 
-iface = gr.Interface(fn=generate_audio, inputs=[input_text, input_emotion], outputs=output_audio)
-iface.launch()
+    app.launch()