Gender Detection through Voice

• PreProcessing
  • Mozilla Common Voice data downloaded from Kaggle
  • Extracted files with Gender attribute
  • Removed extra data from one gender to keep the dataset balanced
  • Processed the data and extracted Mel Spectogram Feature using Librosa
  • Saved all the features in .npy files and all the information in finaldata.csv
• Training
  • Loaded all the data
  • Created a model with 5 fully connected layers with 20% dropout rate
  • Fit the model with Early Stopping with patience 5
  • Saved the model
• Prediction
  • Using mic module from StackOverflow
  • Load the weights to model, predict and return the results
• GUI
  • Created a random grid of colours using HTML and CSS on Flask
  • Added button for turning on the mic
  • Animation using jQuery