Update custom_interface.py

custom_interface.py

@@ -85,64 +85,16 @@ class ASR(Pretrained):
         return seq
     
 
-    # def classify_file(self, path):
-    #     # waveform = self.load_audio(path)
-    #     waveform, sr = librosa.load(path, sr=16000)
-    #     waveform = torch.tensor(waveform)
-
-    #     # Fake a batch:
-    #     batch = waveform.unsqueeze(0)
-    #     rel_length = torch.tensor([1.0])
-    #     outputs = self.encode_batch(batch, rel_length)
-       
-    #     return outputs
-
     def classify_file(self, path):
-        # Load the audio file
+        # waveform = self.load_audio(path)
         waveform, sr = librosa.load(path, sr=16000)
-    
-        # Get audio length in seconds
-        audio_length = len(waveform) / sr
-        print(f"Audio length: {audio_length:.2f} seconds")
-    
-        # Detect non-silent segments
-        non_silent_intervals = librosa.effects.split(waveform, top_db=20)  # Adjust top_db for sensitivity
-    
-        segments = []
-        current_segment = []
-        current_length = 0
-        max_duration = 20 * sr  # Maximum segment duration in samples (20 seconds)
-    
-        for interval in non_silent_intervals:
-            start, end = interval
-            segment_part = waveform[start:end]
-    
-            # If adding the next part exceeds max duration, store the segment and start a new one
-            if current_length + len(segment_part) > max_duration:
-                segments.append(np.concatenate(current_segment))
-                current_segment = []
-                current_length = 0
-    
-            current_segment.append(segment_part)
-            current_length += len(segment_part)
-    
-        # Append the last segment if it's not empty
-        if current_segment:
-            segments.append(np.concatenate(current_segment))
-    
-        # Process each segment
-        outputs = []
-        for i, segment in enumerate(segments):
-            print(f"Processing segment {i + 1}/{len(segments)}, length: {len(segment) / sr:.2f} seconds")
-    
-            segment_tensor = torch.tensor(segment)
-    
-            # Fake a batch for the segment
-            batch = segment_tensor.unsqueeze(0)
-            rel_length = torch.tensor([1.0])  # Adjust if necessary
-    
-            # Pass the segment through the ASR model
-            segment_output = self.encode_batch(batch, rel_length)
-            outputs.append(segment_output)
-    
+        waveform = torch.tensor(waveform)
+
+        # Fake a batch:
+        batch = waveform.unsqueeze(0)
+        rel_length = torch.tensor([1.0])
+        outputs = self.encode_batch(batch, rel_length)
+       
         return outputs
+
+