Update app.py

app.py

@@ -27,13 +27,16 @@ checkpoint_path = load_model()
 
 # 将 detect 函数移到 GPU 装饰器下
 @spaces.GPU
-def detect_on_gpu(dataset):
+def detect_on_gpu(audio_path):
     """在 GPU 上进行音频伪造检测"""
     print("\n=== 开始音频检测 ===")
     
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     print(f"使用设备: {device}")
     
+    # 数据集处理移到GPU函数内部
+    audio_dataset = dataset.DemoDataset(audio_path)
+    
     print("正在初始化模型...")
     model = SpoofVerificationModel().to(device)
     
@@ -59,8 +62,9 @@ def detect_on_gpu(dataset):
 
     print("\n开始处理音频数据...")
     with torch.no_grad():
-        for batch_idx, batch in enumerate(dataset):
+        for batch_idx, batch in enumerate(audio_dataset):
             print(f"\n处理批次 {batch_idx + 1}")
+            print('waveforms shape:', batch['waveforms'].shape)
             waveforms = batch['waveforms'].numpy() # [B, T]
             features = feature_extractor(waveforms, sampling_rate=16000, return_attention_mask=True, padding_value=0, return_tensors="pt").to(device)
             outputs = model(features)
@@ -77,12 +81,9 @@ def detect_on_gpu(dataset):
     return result
 
 def audio_deepfake_detection(audio_path):
-    
-    # 数据集处理
-    audio_dataset = dataset.DemoDataset(audio_path)
-    
-    # 调用 GPU 检测函数
-    result = detect_on_gpu(audio_dataset)
+    # 移除了数据集处理步骤
+    # 直接传递音频路径到GPU函数
+    result = detect_on_gpu(audio_path)
     is_fake = "是/Yes" if result["is_fake"] else "否/No"
     confidence = f"{100*result['confidence']:.2f}%"