Update app.py

app.py

@@ -37,38 +37,34 @@ model.to(device)
 print(f"✅ Model loaded on {device}")
 
 # ================================
-# 3️⃣ Load Dataset (From Extracted Folder)
+# 3️⃣ Load Dataset (With Fixes)
 # ================================
 DATASET_TAR_PATH = "dev-clean.tar.gz"
 EXTRACT_PATH = "./librispeech_dev_clean"
+AUDIO_FOLDER = os.path.join(EXTRACT_PATH, "LibriSpeech", "dev-clean")
 
-if not os.path.exists(EXTRACT_PATH):
+if not os.path.exists(AUDIO_FOLDER):
     print("🔄 Extracting dataset...")
-    with tarfile.open(DATASET_TAR_PATH, "r:gz") as tar:
-        tar.extractall(EXTRACT_PATH)
-    print("✅ Extraction complete.")
+    try:
+        with tarfile.open(DATASET_TAR_PATH, "r:gz") as tar:
+            tar.extractall(EXTRACT_PATH)
+        print("✅ Extraction complete.")
+    except Exception as e:
+        raise RuntimeError(f"❌ Dataset extraction failed: {e}")
 else:
     print("✅ Dataset already extracted.")
 
-AUDIO_FOLDER = os.path.join(EXTRACT_PATH, "LibriSpeech", "dev-clean")
-
 def find_audio_files(base_folder):
-    audio_files = []
-    for root, _, files in os.walk(base_folder):
-        for file in files:
-            if file.endswith(".flac"):
-                audio_files.append(os.path.join(root, file))
-    return audio_files
+    return [os.path.join(root, file) for root, _, files in os.walk(base_folder) for file in files if file.endswith(".flac")]
 
 audio_files = find_audio_files(AUDIO_FOLDER)
 
 if not audio_files:
     raise FileNotFoundError(f"❌ No .flac files found in {AUDIO_FOLDER}. Check dataset structure!")
-
 print(f"✅ Found {len(audio_files)} audio files in dataset!")
 
 # ================================
-# 4️⃣ Load Transcripts
+# 4️⃣ Load Transcripts (Fixed Mapping)
 # ================================
 def load_transcripts():
     transcript_dict = {}
@@ -86,11 +82,10 @@ def load_transcripts():
 transcripts = load_transcripts()
 if not transcripts:
     raise FileNotFoundError("❌ No transcripts found! Check dataset structure.")
-
 print(f"✅ Loaded {len(transcripts)} transcripts.")
 
 # ================================
-# 5️⃣ Preprocess Dataset (Fixing `input_ids` issue)
+# 5️⃣ Preprocess Dataset (Fixed `input_ids` Issue)
 # ================================
 def load_and_process_audio(audio_path):
     waveform, sample_rate = torchaudio.load(audio_path)
@@ -123,9 +118,9 @@ batch_size = st.sidebar.select_slider("Batch Size", options=[2, 4, 8, 16], value
 attack_strength = st.sidebar.slider("Attack Strength", 0.0, 0.9, 0.1)
 
 # ================================
-# 7️⃣ Streamlit ASR Web App (Fast Decoding & Security Features)
+# 7️⃣ Streamlit ASR Web App (Fixed Security & Processing)
 # ================================
-st.title("🎙️ Speech-to-Text ASR Model with Security Features Trained on Libri Speech dataset 🎶")
+st.title("🎙️ Speech-to-Text ASR Model with Security Features 🎶")
 
 audio_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "flac"])
 
@@ -138,9 +133,9 @@ if audio_file:
     waveform = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(waveform)
     waveform = waveform.to(dtype=torch.float32)
     
-    # Simulate an adversarial attack by injecting random noise
-    adversarial_waveform = waveform + (attack_strength * torch.randn_like(waveform))
-    adversarial_waveform = torch.clamp(adversarial_waveform, -1.0, 1.0)
+    # Apply adversarial attack noise with limit
+    noise = torch.randn_like(waveform) * attack_strength
+    adversarial_waveform = torch.clamp(waveform + noise, -1.0, 1.0)
     
     input_features = processor(adversarial_waveform.squeeze().numpy(), sampling_rate=16000, return_tensors="pt").input_features.to(device)