Vocal Extraction from Complex Audio Mixtures

Developed a hybrid model combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to isolate vocals from complex audio mixtures. • Preprocessed the MUSDB18 dataset, applying techniques like spectrogram conversion via Short-Time Fourier Transform (STFT), pitch shifting, time stretching, advanced hyperparameter tuning(Optuna) and dynamic quantization using ONNX format amongst others