sync ms

app.py

@@ -9,18 +9,16 @@ import librosa.display
 import matplotlib.pyplot as plt
 from collections import Counter
 from model import EvalNet
-from utils import get_modelist, find_wav_files, embed_img
-
-
-TRANSLATE = {
-    "m_bel": "Bel Canto, Male",
-    "f_bel": "Bel Canto, Female",
-    "m_folk": "Folk Singing, Male",
-    "f_folk": "Folk Singing, Female",
-}
-CLASSES = list(TRANSLATE.keys())
-TEMP_DIR = "./__pycache__/tmp"
-SAMPLE_RATE = 22050
+from utils import (
+    get_modelist,
+    find_wav_files,
+    embed_img,
+    _L,
+    SAMPLE_RATE,
+    TEMP_DIR,
+    TRANSLATE,
+    CLASSES,
+)
 
 
 def wav2mel(audio_path: str, width=1.6, topdb=40):
@@ -99,34 +97,38 @@ def most_common_element(input_list: list):
 
 
 def infer(wav_path: str, log_name: str, folder_path=TEMP_DIR):
-    if os.path.exists(folder_path):
-        shutil.rmtree(folder_path)
+    status = "Success"
+    filename = result = None
+    try:
+        if os.path.exists(folder_path):
+            shutil.rmtree(folder_path)
 
-    if not wav_path:
-        return None, "Please input an audio!"
+        if not wav_path:
+            raise ValueError("请输入音频!")
 
-    spec = log_name.split("_")[-3]
-    os.makedirs(folder_path, exist_ok=True)
-    try:
+        spec = log_name.split("_")[-3]
+        os.makedirs(folder_path, exist_ok=True)
         model = EvalNet(log_name, len(TRANSLATE)).model
         eval("wav2%s" % spec)(wav_path)
+        outputs = []
+        all_files = os.listdir(folder_path)
+        for file_name in all_files:
+            if file_name.lower().endswith(".jpg"):
+                file_path = os.path.join(folder_path, file_name)
+                input = embed_img(file_path)
+                output: torch.Tensor = model(input)
+                pred_id = torch.max(output.data, 1)[1]
+                outputs.append(int(pred_id))
+
+        max_count_item = most_common_element(outputs)
+        shutil.rmtree(folder_path)
+        filename = os.path.basename(wav_path)
+        result = TRANSLATE[CLASSES[max_count_item]]
 
     except Exception as e:
-        return None, f"{e}"
-
-    outputs = []
-    all_files = os.listdir(folder_path)
-    for file_name in all_files:
-        if file_name.lower().endswith(".jpg"):
-            file_path = os.path.join(folder_path, file_name)
-            input = embed_img(file_path)
-            output: torch.Tensor = model(input)
-            pred_id = torch.max(output.data, 1)[1]
-            outputs.append(int(pred_id))
+        status = f"{e}"
 
-    max_count_item = most_common_element(outputs)
-    shutil.rmtree(folder_path)
-    return os.path.basename(wav_path), TRANSLATE[CLASSES[max_count_item]]
+    return status, filename, result
 
 
 if __name__ == "__main__":
@@ -141,36 +143,37 @@ if __name__ == "__main__":
         gr.Interface(
             fn=infer,
             inputs=[
-                gr.Audio(label="Upload a recording (>40dB)", type="filepath"),
-                gr.Dropdown(choices=models, label="Select a model", value=models[0]),
+                gr.Audio(label=_L("上传录音 (>40dB)"), type="filepath"),
+                gr.Dropdown(choices=models, label=_L("选择模型"), value=models[0]),
             ],
             outputs=[
-                gr.Textbox(label="Audio filename", show_copy_button=True),
-                gr.Textbox(label="Singing method recognition", show_copy_button=True),
+                gr.Textbox(label=_L("状态栏"), show_copy_button=True),
+                gr.Textbox(label=_L("音频文件名"), show_copy_button=True),
+                gr.Textbox(label=_L("唱法识别"), show_copy_button=True),
             ],
             examples=examples,
             cache_examples=False,
             allow_flagging="never",
-            title="It is recommended to keep the recording length around 5s, too long will affect the recognition efficiency.",
+            title=_L("建议录音时长保持在 5s 左右, 过长会影响识别效率"),
         )
 
         gr.Markdown(
-            """
-# Cite
-```bibtex
-@article{Zhou-2025,
-  author  = {Monan Zhou and Shenyang Xu and Zhaorui Liu and Zhaowen Wang and Feng Yu and Wei Li and Baoqiang Han},
-  title   = {CCMusic: An Open and Diverse Database for Chinese Music Information Retrieval Research},
-  journal = {Transactions of the International Society for Music Information Retrieval},
-  volume  = {8},
-  number  = {1},
-  pages   = {22--38},
-  month   = {Mar},
-  year    = {2025},
-  url     = {https://doi.org/10.5334/tismir.194},
-  doi     = {10.5334/tismir.194}
-}
-```"""
+            f"# {_L('引用')}"
+            + """
+            ```bibtex
+            @article{Zhou-2025,
+                author  = {Monan Zhou and Shenyang Xu and Zhaorui Liu and Zhaowen Wang and Feng Yu and Wei Li and Baoqiang Han},
+                title   = {CCMusic: An Open and Diverse Database for Chinese Music Information Retrieval Research},
+                journal = {Transactions of the International Society for Music Information Retrieval},
+                volume  = {8},
+                number  = {1},
+                pages   = {22--38},
+                month   = {Mar},
+                year    = {2025},
+                url     = {https://doi.org/10.5334/tismir.194},
+                doi     = {10.5334/tismir.194}
+            }
+            ```"""
         )
 
     demo.launch()

requirements.txt

@@ -1,5 +1,7 @@
-torch
-pillow
+torch==2.6.0+cu118
+-f https://download.pytorch.org/whl/torch
+torchvision==0.21.0+cu118
+-f https://download.pytorch.org/whl/torchvision
 librosa
 matplotlib
-torchvision
+modelscope[framework]==1.21.0

utils.py

@@ -1,10 +1,53 @@
 import os
 import torch
 import torchvision.transforms as transforms
-from huggingface_hub import snapshot_download
+import huggingface_hub
+import modelscope
 from PIL import Image
 
-MODEL_DIR = snapshot_download("ccmusic-database/bel_canto", cache_dir="./__pycache__")
+
+EN_US = os.getenv("LANG") != "zh_CN.UTF-8"
+
+ZH2EN = {
+    "上传录音 (>40dB)": "Upload a recording (>40dB)",
+    "选择模型": "Select a model",
+    "状态栏": "Status",
+    "音频文件名": "Audio filename",
+    "唱法识别": "Singing method recognition",
+    "建议录音时长保持在 5s 左右, 过长会影响识别效率": "It is recommended to keep the recording length around 5s, too long will affect the recognition efficiency.",
+    "引用": "Cite",
+    "男声 & 美声唱法": "Bel Canto, Male",
+    "女声 & 美声唱法": "Bel Canto, Female",
+    "男声 & 民族唱法": "Folk Singing, Male",
+    "女声 & 民族唱法": "Folk Singing, Female",
+}
+
+MODEL_DIR = (
+    huggingface_hub.snapshot_download(
+        "ccmusic-database/bel_canto",
+        cache_dir="./__pycache__",
+    )
+    if EN_US
+    else modelscope.snapshot_download(
+        "ccmusic-database/bel_canto",
+        cache_dir="./__pycache__",
+    )
+)
+
+
+def _L(zh_txt: str):
+    return ZH2EN[zh_txt] if EN_US else zh_txt
+
+
+TRANSLATE = {
+    "m_bel": _L("男声 & 美声唱法"),
+    "f_bel": _L("女声 & 美声唱法"),
+    "m_folk": _L("男声 & 民族唱法"),
+    "f_folk": _L("女声 & 民族唱法"),
+}
+CLASSES = list(TRANSLATE.keys())
+TEMP_DIR = "./__pycache__/tmp"
+SAMPLE_RATE = 22050
 
 
 def toCUDA(x):
@@ -27,19 +70,16 @@ def find_wav_files(folder_path=f"{MODEL_DIR}/examples"):
 
 
 def get_modelist(model_dir=MODEL_DIR, assign_model=""):
-    try:
-        entries = os.listdir(model_dir)
-    except OSError as e:
-        print(f"Cannot access {model_dir}: {e}")
-        return
-
     output = []
-    for entry in entries:
+    for entry in os.listdir(model_dir):
+        # 获取完整路径
         full_path = os.path.join(model_dir, entry)
+        # 跳过'.git'文件夹
         if entry == ".git" or entry == "examples":
-            print(f"Skip .git / examples dir: {full_path}")
+            print(f"跳过 .git 或 examples 文件夹: {full_path}")
             continue
 
+        # 检查条目是文件还是目录
         if os.path.isdir(full_path):
             model = os.path.basename(full_path)
             if assign_model and assign_model.lower() in model: