Update app.py

app.py

@@ -0,0 +1,174 @@
+import gradio as gr
+import torch
+import torch.nn as nn
+import librosa
+import numpy as np
+import whisper
+import pandas as pd
+from datasets import load_dataset
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.linear_model import LogisticRegression
+
+device = torch.device("cpu")
+
+# ================= LOAD DATASET =================
+data1 = pd.read_csv("spam_dataset.csv")
+data2 = load_dataset("ucirvine/sms_spam")
+
+# convert to pandas
+data2 = data2["train"].to_pandas()
+
+# rename columns to match your dataset
+data2 = data2.rename(columns={"sms": "text", "label": "label"})
+
+# combine both
+data = pd.concat([data1, data2], ignore_index=True)
+
+texts = data["text"]
+labels = data["label"]
+
+# ================= ML TRAINING =================
+vectorizer = TfidfVectorizer()
+X = vectorizer.fit_transform(texts)
+
+ml_model = LogisticRegression()
+ml_model.fit(X, labels)
+
+# ================= CNN MODEL =================
+class ScamAudioCNN(nn.Module):
+    def __init__(self):
+        super(ScamAudioCNN, self).__init__()
+        self.conv1 = nn.Conv2d(1, 16, 3, padding=1)
+        self.pool = nn.MaxPool2d(2, 2)
+        self.conv2 = nn.Conv2d(16, 32, 3, padding=1)
+        self.fc1 = nn.Linear(32 * 10 * 25, 128)
+        self.fc2 = nn.Linear(128, 2)
+
+    def forward(self, x):
+        x = self.pool(torch.relu(self.conv1(x)))
+        x = self.pool(torch.relu(self.conv2(x)))
+        x = x.view(x.size(0), -1)
+        x = torch.relu(self.fc1(x))
+        x = self.fc2(x)
+        return x
+
+cnn_model = ScamAudioCNN().to(device)
+
+# ================= LOAD CNN (optional) =================
+try:
+    cnn_model.load_state_dict(torch.load("scam_audio_model.pth", map_location=device))
+except:
+    pass
+
+cnn_model.eval()
+
+# ================= WHISPER =================
+whisper_model = whisper.load_model("tiny", device="cpu")
+
+# ================= MFCC =================
+def extract_features(file_path, max_len=100):
+    y, sr = librosa.load(file_path, sr=16000)
+
+    mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40)
+
+    if mfcc.shape[1] < max_len:
+        mfcc = np.pad(mfcc, ((0,0),(0,max_len-mfcc.shape[1])))
+    else:
+        mfcc = mfcc[:, :max_len]
+
+    mfcc = mfcc[np.newaxis, np.newaxis, :, :]
+    return torch.tensor(mfcc, dtype=torch.float32)
+
+# ================= TRANSCRIPTION =================
+def transcribe_audio(file_path):
+    result = whisper_model.transcribe(file_path)
+    return result["text"].lower()
+
+# ================= KEYWORDS =================
+scam_keywords = [
+    "otp","bank","account","verify","urgent","blocked","suspend",
+    "credit card","loan","refund","investment","crypto","kyc",
+    "password","security","congratulations","won","winner","prize",
+    "claim","fee","pay","offer","lottery","jackpot","gift","free"
+]
+
+def keyword_score(text):
+    found = [w for w in scam_keywords if w in text]
+    score = min(len(found)/4, 1.0)
+    return score, found
+
+# ================= ML PREDICTION =================
+def ml_predict(text):
+    X_test = vectorizer.transform([text])
+    prob = ml_model.predict_proba(X_test)[0][1]
+    return prob
+
+# ================= MAIN =================
+def analyze_audio(audio):
+
+    if audio is None:
+        return "No audio detected."
+
+    try:
+        # Whisper
+        transcript = transcribe_audio(audio)
+
+        # Keyword
+        k_score, words = keyword_score(transcript)
+
+        # ML
+        ml_score = ml_predict(transcript)
+
+        # CNN
+        features = extract_features(audio).to(device)
+        with torch.no_grad():
+            out = cnn_model(features)
+            probs = torch.softmax(out, dim=1)
+            cnn_score = probs[0][1].item()
+
+        # FINAL FUSION
+        final_score = (0.3 * k_score) + (0.4 * ml_score) + (0.3 * cnn_score)
+
+        if final_score < 0.30:
+            risk = "Low Risk"
+        elif final_score < 0.60:
+            risk = "Medium Risk"
+        else:
+            risk = "High Scam Risk"
+
+        result = "SPAM" if final_score >= 0.30 else "NOT SPAM"
+
+        return f"""
+Transcript: {transcript}
+
+Spam Words Found: {', '.join(words) if words else 'None'}
+
+Keyword Score: {k_score:.2f}
+ML Score: {ml_score:.2f}
+CNN Score: {cnn_score:.2f}
+
+Final Probability: {final_score*100:.2f}%
+
+Risk Level: {risk}
+
+Final Result: {result}
+"""
+
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+# ================= UI =================
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎙️ Hybrid Voice Scam Detection System")
+    gr.Markdown("Using speech and content analysis")
+    
+    audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath")
+    output = gr.Textbox()
+    
+    gr.Button("Analyze").click(
+        analyze_audio,
+        inputs=audio_input,
+        outputs=output
+    )
+
+demo.launch()