Hyperparameter Optimization of CNN Using Genetic Algorithm for Speech Command Recognition

Abstract

Identification or recognition of speaker uttered short speech commands, known as speech command recognition (SCR), is a challenging problem in digital speech processing. It plays a vital role in various real-life applications such as Internet of things (IoT) devices and assistive technology to name a few. The deep learning models like convolutional neural networks (CNNs) have shown the potential ability to solve SCR tasks. However, these models’ performance heavily depends on the appropriate choice of associated hyperparameters values, which are very labor-intensive and time-consuming. This paper proposes a genetic algorithm (GA)-based hyperparameters optimization approach to fine-tune CNN hyperparameters for performing SCR tasks. The proposed method is trained and tested with the google speech command (GSC) dataset by considering eight speech commands. The experimental results demonstrate the efficiency and effectiveness of the proposed method for solving speech features-based classification problems like SCR.