app.py

import gradio as gr
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchaudio
import os

# ----------------------------
# Define Model
# ----------------------------
class AudioCNN(nn.Module):
    def __init__(self, num_classes=3):
        super(AudioCNN, self).__init__()
        self.conv1 = nn.Conv1d(1, 16, kernel_size=5, stride=2)
        self.pool = nn.MaxPool1d(2)
        self.global_pool = nn.AdaptiveAvgPool1d(1)
        self.fc1 = nn.Linear(16, num_classes)

    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))  # [B, 16, L']
        x = self.global_pool(x)                # [B, 16, 1]
        x = x.view(x.size(0), -1)             # [B, 16]
        x = self.fc1(x)                       # [B, num_classes]
        return x

# ----------------------------
# Load model
# ----------------------------
num_classes = 3
model_save_path = "audio_cnn_model.pth"

model = AudioCNN(num_classes)

if os.path.exists(model_save_path):
    try:
        model.load_state_dict(torch.load(model_save_path, map_location=torch.device('cpu')), strict=False)
        model.eval()
        print(f"✅ Model state dictionary loaded from {model_save_path}")
    except Exception as e:
        print(f"⚠️ Error loading model state dictionary: {e}")
        model = None
else:
    print(f"⚠️ Model state dictionary not found at {model_save_path}")
    model = None

# ----------------------------
# Prediction function
# ----------------------------
def predict_audio(audio_file_path):
    if model is None:
        return "Model not loaded. Cannot make predictions."
    if audio_file_path is None:
        return "No audio file provided."

    try:
        waveform, sample_rate = torchaudio.load(audio_file_path)

        # Convert stereo → mono
        if waveform.shape[0] > 1:
            waveform = torch.mean(waveform, dim=0, keepdim=True)

        # Normalize waveform
        waveform = (waveform - waveform.mean()) / (waveform.std() + 1e-6)

        # Ensure correct shape [batch, channels, length]
        waveform = waveform.unsqueeze(0)

        # Predict
        with torch.no_grad():
            outputs = model(waveform)
        _, predicted_index = torch.max(outputs.data, 1)
        predicted_index = predicted_index.item()

        label_map = {0: 'English', 1: 'Code-switched', 2: 'Other'}
        predicted_label = label_map.get(predicted_index, "Unknown")
        return predicted_label

    except Exception as e:
        return f"Error during prediction: {e}"

# ----------------------------
# Launch Gradio
# ----------------------------
if model is not None:
    interface = gr.Interface(
        fn=predict_audio,
        inputs=gr.Audio(type="filepath"),
        outputs=gr.Label(),
        title="Audio Code-Switching Detector",
        description="Upload an audio file to detect if it contains code-switching."
    )
    interface.launch(share=True)
else:
    print("⚠️ Gradio interface not created due to model loading error.")