修改了拼接的方法

api.py

@@ -215,14 +215,14 @@ async def process_audio(audio_data: bytes, language: str = "auto") -> str:
             resampler = torchaudio.transforms.Resample(sample_rate, 16000)
             input_wav = resampler(torch.from_numpy(input_wav)[None, :])[0, :].numpy()
         
-        target_length = 90 * sample_rate 
-        current_length = input_wav.shape[0]
+        print(input_wav.shape)
+
+        target_length = 90 * 16000 
+        current_length = input_wav.shape[1]
         if current_length < target_length:
             padding_length = target_length - current_length
-            padding = np.zeros(padding_length, dtype=np.float32)
+            padding = np.zeros((1, padding_length), dtype=np.float32)
             input_wav = np.concatenate((input_wav, padding))
-        elif current_length > target_length:
-            input_wav = input_wav[:target_length]
 
         # Model inference
         text = model.generate(