Pin gradio version

app.py

@@ -1,20 +1,12 @@
 import os
+os.system("pip install gradio==2.8.0b2")
 import gradio as gr
 import torch
-import pydub
-import torchaudio
-from torchaudio.sox_effects import apply_effects_tensor
-import numpy as np
+from torchaudio.sox_effects import apply_effects_file
 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.astype(np.float32), audio.frame_rate
-
 STYLE = """
 <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" integrity="sha256-YvdLHPgkqJ8DVUxjjnGVlMMJtNimJ6dYkowFFvp4kKs=" crossorigin="anonymous">
 """
@@ -63,12 +55,9 @@ cosine_sim = torch.nn.CosineSimilarity(dim=-1)
 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 = load_audio(path1)
-    print(wav1, wav1.shape, wav1.dtype)
-    wav1, _ = apply_effects_tensor(torch.tensor(wav1).unsqueeze(0), sr1, EFFECTS)
-    wav2, sr2 = load_audio(path2)
-    wav2, _ = apply_effects_tensor(torch.tensor(wav2).unsqueeze(0), sr2, EFFECTS)
+
+    wav1, _ = apply_effects_file(path1, EFFECTS)
+    wav2, _ = apply_effects_file(path2, EFFECTS)
     print(wav1.shape, wav2.shape)
 
     input1 = feature_extractor(wav1.squeeze(0), return_tensors="pt", sampling_rate=16000).input_values.to(device)
@@ -138,4 +127,4 @@ interface = gr.Interface(
     live=False,
     examples=examples,
 )
-interface.launch(enable_queue=True)
+interface.launch(enable_queue=True)