Update app.py

app.py

@@ -72,15 +72,15 @@ from io import BytesIO
 
 # Update the function to handle both file paths and Streamlit-uploaded files
 def get_audio_embedding(file_input):
-    # Determine if input is a file path (string) or a file-like object (uploaded file)
-    if isinstance(file_input, str):
-        # If the input is a file path, use it directly with librosa
-        audio, sr = librosa.load(file_input, sr=16000)
-    else:
-        # If the input is an uploaded file (BytesIO), use soundfile to read it
-        audio, sr = sf.read(BytesIO(file_input.read()), dtype='float32')
-
-    # Extract features using the Wav2Vec2 model
+    # If the input is a Streamlit-uploaded file, convert it to bytes and then load using librosa
+    if not isinstance(file_input, str):
+        # Streamlit uploads the file as an in-memory file-like object, so we convert it into bytes
+        file_input = BytesIO(file_input.read())
+    
+    # Load the audio using librosa
+    audio, sr = librosa.load(file_input, sr=16000)
+    
+    # Convert audio to embeddings using Wav2Vec2
     inputs = feature_extractor(audio, sampling_rate=sr, return_tensors="pt", padding=True)
     with torch.no_grad():
         embeddings = model(**inputs).last_hidden_state.mean(dim=1)