sync ms

app.py

@@ -1,19 +1,18 @@
-import re
-import os
-import sys
+from text.symbols import symbols
+from text.cleaner import clean_text
+from text import cleaned_text_to_sequence, get_bert
+from models import SynthesizerTrn
+from tqdm import tqdm
+from utils import _L, MODEL_DIR
+import gradio as gr
+import numpy as np
+import commons
+import random
 import utils
 import torch
-import random
-import commons
-import numpy as np
-import gradio as gr
-from tqdm import tqdm
-from models import SynthesizerTrn
-from huggingface_hub import snapshot_download
-from text import cleaned_text_to_sequence, get_bert
-from text.cleaner import clean_text
-from text.symbols import symbols
-
+import sys
+import re
+import os
 
 if sys.platform == "darwin":
     os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
@@ -25,7 +24,8 @@ logging.getLogger("markdown_it").setLevel(logging.WARNING)
 logging.getLogger("urllib3").setLevel(logging.WARNING)
 logging.getLogger("matplotlib").setLevel(logging.WARNING)
 logging.basicConfig(
-    level=logging.INFO, format="| %(name)s | %(levelname)s | %(message)s"
+    level=logging.INFO,
+    format="| %(name)s | %(levelname)s | %(message)s",
 )
 
 logger = logging.getLogger(__name__)
@@ -102,34 +102,39 @@ def tts_fn(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale):
 
 def text_splitter(text: str):
     punctuation = r"[。,；,！,？,〜,\n,\r,\t,.,!,;,?,~, ]"
+    # 使用正则表达式根据标点符号分割文本, 并忽略重叠的分隔符
     sentences = re.split(punctuation, text.strip())
+    # 过滤掉空字符串
     return [sentence.strip() for sentence in sentences if sentence.strip()]
 
 
 def concatenate_audios(audio_samples, sample_rate=44100):
     half_second_silence = np.zeros(int(sample_rate / 2))
+    # 初始化最终的音频数组
     final_audio = audio_samples[0]
+    # 遍历音频样本列表, 并将它们连接起来, 每个样本之间插入半秒钟的静音
     for sample in audio_samples[1:]:
         final_audio = np.concatenate((final_audio, half_second_silence, sample))
 
-    print("Audio pieces concatenated!")
+    print("音频片段连接完成！")
     return (sample_rate, final_audio)
 
 
 def read_text(file_path: str):
     try:
+        # 打开文件并读取内容
         with open(file_path, "r", encoding="utf-8") as file:
             content = file.read()
             return content
 
     except FileNotFoundError:
-        print(f"File Not Found: {file_path}")
+        print(f"文件未找到: {file_path}")
 
     except IOError:
-        print(f"An error occurred reading the file: {file_path}")
+        print(f"读取文件时发生错误: {file_path}")
 
     except Exception as e:
-        print(f"An unknown error has occurred: {e}")
+        print(f"发生未知错误: {e}")
 
 
 def infer_tab1(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale):
@@ -137,7 +142,7 @@ def infer_tab1(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scal
         content = read_text(text)
         sentences = text_splitter(content)
         audios = []
-        for sentence in tqdm(sentences, desc="TTS inferring..."):
+        for sentence in tqdm(sentences, desc="TTS 推理中..."):
             with torch.no_grad():
                 audios.append(
                     infer(
@@ -160,7 +165,7 @@ def infer_tab2(content, speaker, sdp_ratio, noise_scale, noise_scale_w, length_s
     try:
         sentences = text_splitter(content)
         audios = []
-        for sentence in tqdm(sentences, desc="TTS inferring..."):
+        for sentence in tqdm(sentences, desc="TTS 推理中..."):
             with torch.no_grad():
                 audios.append(
                     infer(
@@ -181,12 +186,11 @@ def infer_tab2(content, speaker, sdp_ratio, noise_scale, noise_scale_w, length_s
 
 
 if __name__ == "__main__":
-    model_dir = snapshot_download("Genius-Society/hoyoTTS", cache_dir="./__pycache__")
     if debug:
         logger.info("Enable DEBUG-LEVEL log")
         logging.basicConfig(level=logging.DEBUG)
 
-    hps = utils.get_hparams_from_dir(model_dir)
+    hps = utils.get_hparams_from_dir(MODEL_DIR)
     device = (
         "cuda:0"
         if torch.cuda.is_available()
@@ -204,105 +208,70 @@ if __name__ == "__main__":
         **hps.model,
     ).to(device)
     net_g.eval()
-    utils.load_checkpoint(f"{model_dir}/G_78000.pth", net_g, None, skip_optimizer=True)
+    utils.load_checkpoint(f"{MODEL_DIR}/G_78000.pth", net_g, None, skip_optimizer=True)
     speaker_ids = hps.data.spk2id
     speakers = list(speaker_ids.keys())
     random.shuffle(speakers)
     with gr.Blocks() as app:
         gr.Markdown(
             """
-Welcome to the Space, which is based on the open source project <a href="https://github.com/fishaudio/Bert-VITS2">Bert-vits2</a>, and moved to the bottom for an explanation of the principle. This Space must be used in accordance with local laws and regulations, prohibiting the use of it for any criminal activities."""
+欢迎使用此创空间，此创空间基于 <a href="https://github.com/fishaudio/Bert-VITS2">Bert-vits2</a> 开源项目制作，移至最底端有原理浅讲。使用此创空间必须遵守当地相关法律法规，禁止用其从事任何违法犯罪活动。"""
         )
 
-        with gr.Tab("Input Mode"):
+        with gr.Tab("输入模式"):
             gr.Interface(
-                fn=infer_tab2,
+                fn=infer_tab2,  # 使用 text_to_speech 函数
                 inputs=[
                     gr.TextArea(
-                        label="Please input the Simplified Chinese text",
-                        placeholder="The first inference takes time to download the model, so be patient.",
+                        label="请输入简体中文文案",
+                        placeholder="首次推理需耗时下载模型，还请耐心等待。",
                         show_copy_button=True,
                     ),
-                    gr.Dropdown(choices=speakers, value="莱依拉", label="Role"),
+                    gr.Dropdown(choices=speakers, value="莱依拉", label="角色"),
                     gr.Slider(
-                        minimum=0,
-                        maximum=1,
-                        value=0.2,
-                        step=0.1,
-                        label="Modulation of intonation",
-                    ),  # SDP/DP Mix Ratio
+                        minimum=0, maximum=1, value=0.2, step=0.1, label="语调调节"
+                    ),  # SDP/DP混合比
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=0.6,
-                        step=0.1,
-                        label="Emotional adjustment",
+                        minimum=0.1, maximum=2, value=0.6, step=0.1, label="感情调节"
                     ),
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=0.8,
-                        step=0.1,
-                        label="Phoneme length",
+                        minimum=0.1, maximum=2, value=0.8, step=0.1, label="音素长度"
                     ),
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=1,
-                        step=0.1,
-                        label="Output duration",
+                        minimum=0.1, maximum=2, value=1, step=0.1, label="生成时长"
                     ),
                 ],
-                outputs=gr.Audio(label="Output Audio", show_share_button=False),
+                outputs=gr.Audio(label="输出音频"),
                 flagging_mode="never",
                 concurrency_limit=4,
             )
 
-        with gr.Tab("Upload Mode"):
+        with gr.Tab("上传模式"):
             gr.Interface(
-                fn=infer_tab1,  # Use text_to_speech func
+                fn=infer_tab1,  # 使用 text_to_speech 函数
                 inputs=[
                     gr.components.File(
-                        label="Please upload a simplified Chinese TXT",
+                        label="请上传简体中文 TXT 文案",
                         type="filepath",
                         file_types=[".txt"],
                     ),
-                    gr.Dropdown(choices=speakers, value="莱依拉", label="Role"),
+                    gr.Dropdown(choices=speakers, value="莱依拉", label="角色"),
                     gr.Slider(
-                        minimum=0,
-                        maximum=1,
-                        value=0.2,
-                        step=0.1,
-                        label="Modulation of intonation",
-                    ),
+                        minimum=0, maximum=1, value=0.2, step=0.1, label="语调调节"
+                    ),  # SDP/DP混合比
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=0.6,
-                        step=0.1,
-                        label="Emotional adjustment",
+                        minimum=0.1, maximum=2, value=0.6, step=0.1, label="感情调节"
                     ),
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=0.8,
-                        step=0.1,
-                        label="Phoneme length",
+                        minimum=0.1, maximum=2, value=0.8, step=0.1, label="音素长度"
                     ),
                     gr.Slider(
-                        minimum=0.1,
-                        maximum=2,
-                        value=1,
-                        step=0.1,
-                        label="Output duration",
+                        minimum=0.1, maximum=2, value=1, step=0.1, label="生成时长"
                     ),
                 ],
                 outputs=[
-                    gr.Audio(label="Output Audio", show_share_button=False),
-                    gr.TextArea(
-                        label="Result of TXT extraction",
-                        show_copy_button=True,
-                    ),
+                    gr.Audio(label="输出音频"),
+                    gr.TextArea(label="文案提取结果", show_copy_button=True),
                 ],
                 flagging_mode="never",
                 concurrency_limit=4,

requirements.txt

@@ -1,9 +1,11 @@
+torch==2.6.0+cu118
+-f https://download.pytorch.org/whl/torch
 av
 cn2an
 jieba
 numba
 scipy
-gradio
+librosa
 pypinyin
 Unidecode
 matplotlib
@@ -11,6 +13,4 @@ phonemizer
 tensorboard
 amfm_decompy
 transformers
-torch==2.3.1
-numpy==1.26.4
-librosa==0.9.1
+numpy==1.26.4

text/chinese_bert.py

@@ -1,6 +1,8 @@
 import sys
 import torch
+from modelscope import snapshot_download
 from transformers import AutoTokenizer, AutoModelForMaskedLM
+from utils import EN_US
 
 device = torch.device(
     "cuda"
@@ -13,7 +15,11 @@ device = torch.device(
 )
 
 # 模型下载
-model_dir = "hfl/chinese-roberta-wwm-ext-large"
+model_dir = (
+    "hfl/chinese-roberta-wwm-ext-large"
+    if EN_US
+    else snapshot_download("dienstag/chinese-roberta-wwm-ext-large")
+)
 tokenizer = AutoTokenizer.from_pretrained(model_dir)
 model = AutoModelForMaskedLM.from_pretrained(model_dir).to(device)
 

utils.py

@@ -6,14 +6,50 @@ import logging
 import argparse
 import requests
 import subprocess
+import modelscope
+import huggingface_hub
 import numpy as np
 from tqdm import tqdm
 from scipy.io.wavfile import read
 
 
 MATPLOTLIB_FLAG = False
-
 logger = logging.getLogger(__name__)
+EN_US = os.getenv("LANG") != "zh_CN.UTF-8"
+ZH2EN = {
+    "输入模式": "Input Mode",
+    "请输入简体中文文案": "Please input the Simplified Chinese text",
+    "首次推理需耗时下载模型，还请耐心等待。": "The first inference takes time to download the model, so be patient.",
+    "角色": "Role",
+    "状态栏": "Status",
+    "语调调节": "Modulation of intonation",
+    "感情调节": "Emotional adjustment",
+    "音素长度": "Phoneme length",
+    "生成时长": "Output duration",
+    "输出音频": "Output Audio",
+    "上传模式": "Upload Mode",
+    "请上传简体中文 TXT 文案": "Please upload a simplified Chinese TXT",
+    "文案提取结果": "Result of TXT extraction",
+    """
+欢迎使用此创空间，此创空间基于 <a href="https://github.com/fishaudio/Bert-VITS2">Bert-vits2</a> 开源项目制作，移至最底端有原理浅讲。使用此创空间必须遵守当地相关法律法规，禁止用其从事任何违法犯罪活动。""": """
+Welcome to the Space, which is based on the open source project <a href="https://github.com/fishaudio/Bert-VITS2">Bert-vits2</a>, and moved to the bottom for an explanation of the principle. This Space must be used in accordance with local laws and regulations, prohibiting the use of it for any criminal activities.""",
+}
+
+MODEL_DIR = (
+    huggingface_hub.snapshot_download(
+        "Genius-Society/hoyoTTS",
+        cache_dir="./__pycache__",
+    )
+    if EN_US
+    else modelscope.snapshot_download(
+        "Genius-Society/hoyoTTS",
+        cache_dir="./__pycache__",
+    )
+)
+
+
+def _L(zh_txt: str):
+    return ZH2EN[zh_txt] if EN_US else zh_txt
 
 
 def load_checkpoint(checkpoint_path, model, optimizer=None, skip_optimizer=False):