DefinePK

Abstract

In the world of artificial intelligence and speech technology, it's becoming increasingly crucial to improve how we filter out background noise from audio, aiming for efficiency without unnecessary complexity. So, the challenge is to come up with a really effective algorithm for real-time noise reduction, ensuring optimal performance. In this study, we've delved into a deep learning approach using a convolutional neural network (CNN) to tackle noise in audio signals. We trained our model on a substantial dataset named "Edinburgh DataShare". Throughout the development of the CNN model, we incorporated Softmax and rectified linear unit as an activation functions, along with the ADAM optimization algorithm. To model evaluation, the model over 50 epochs showed a really low loss of 0.012. Hence, our findings affirm that the CNN network performs well in effectively mitigating noise from audio signals.