Update app.py

app.py

@@ -1,13 +1,20 @@
 import os
 import gradio as gr
 import torch
-import librosa
+import pydub
 import torchaudio
 from torchaudio.sox_effects import apply_effects_tensor
+import numpy as np
 from transformers import AutoFeatureExtractor, AutoModelForAudioXVector
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
+def load_audio(file_name):
+    audio = pydub.AudioSegment.from_file(file_name)
+    arr = np.array(audio.get_array_of_samples(), dtype=np.float32)
+    arr = arr / (1 << (8 * audio.sample_width - 1))
+    return arr, audio.frame_rate
+
 STYLE = """
 <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bootstrap@5.1.3/dist/css/bootstrap.min.css" integrity="sha256-YvdLHPgkqJ8DVUxjjnGVlMMJtNimJ6dYkowFFvp4kKs=" crossorigin="anonymous">
 """
@@ -57,9 +64,9 @@ def similarity_fn(path1, path2):
     if not (path1 and path2):
         return '<b style="color:red">ERROR: Please record audio for *both* speakers!</b>'
     
-    wav1, sr1 = librosa.load(path1, mono=True)
+    wav1, sr1 = load_audio(path1)
     wav1, _ = apply_effects_tensor(torch.tensor(wav1).unsqueeze(0), sr1, EFFECTS)
-    wav2, sr2 = librosa.load(path2, mono=True)
+    wav2, sr2 = load_audio(path2)
     wav2, _ = apply_effects_tensor(torch.tensor(wav2).unsqueeze(0), sr2, EFFECTS)
     print(wav1.shape, wav2.shape)