Update app.py

app.py

@@ -4,7 +4,7 @@ import torch
 import soundfile as sf
 import numpy as np
 import gradio as gr
-
+import librosa
 
 class HuggingFaceFeatureExtractor:
     def __init__(self, model_class, name):
@@ -26,7 +26,6 @@ class HuggingFaceFeatureExtractor:
             outputs = self.model(**inputs)
         return outputs.last_hidden_state
 
-
 FEATURE_EXTRACTORS = {
     "wavlm-base": lambda: HuggingFaceFeatureExtractor(WavLMModel, "microsoft/wavlm-base"),
     "wavLM-V1": lambda: HuggingFaceFeatureExtractor(WavLMModel, "DavidCombei/wavLM-base-DeepFake_UTCN"),
@@ -34,23 +33,17 @@ FEATURE_EXTRACTORS = {
     "wavLM-V3": lambda: HuggingFaceFeatureExtractor(WavLMModel, "DavidCombei/wavLM-base-UTCN_114k"),
 }
 
-
 model1 = joblib.load('model1.joblib')
 model2 = joblib.load('model2.joblib')
 model3 = joblib.load('model3.joblib')
 model4 = joblib.load('model4.joblib')
 final_model = joblib.load('final_model.joblib')
 
-
 def process_audio(file_audio):
-    #audio, sr = sf.read(file_audio)
-        audio, sr = librosa.load(file_audio,sr=16000)
-
+    audio, sr = librosa.load(file_audio, sr=16000)  # Resample to 16 kHz
 
-    if len(audio.shape)>1:
-       audio = audio[0]
-        
-        
+    if len(audio.shape) > 1:
+        audio = audio[0]
 
     extractor_1 = FEATURE_EXTRACTORS['wavlm-base']()
     extractor_2 = FEATURE_EXTRACTORS['wavLM-V1']()
@@ -84,17 +77,16 @@ def process_audio(file_audio):
     final_prob = final_model.predict_proba(eval_combined_probs)[:, 1]
 
     if final_prob < 0.5:
-        return f"Fake with a confidence of: {100-final_prob[0] * 100:.2f}"
+        return f"Fake with a confidence of: {100 - final_prob[0] * 100:.2f}%"
     else:
-        return f"Real with a confidence of: {final_prob[0] * 100:.2f}"
-
+        return f"Real with a confidence of: {final_prob[0] * 100:.2f}%"
 
 interface = gr.Interface(
     fn=process_audio,
     inputs=gr.Audio(type="filepath"),
     outputs="text",
     title="Audio Deepfake Detection",
-    description="Upload an audio file to detect whether it is fake or real.",
+    description="Upload an audio file to detect whether it is fake or real. The system uses features ensamble from wavLM base and finetuned versions. Submitted to ASVSpoof5.",
 )
 
 interface.launch(share=True)