Arwin Datumaya Wahyudi Sumari, Zest Amborgang Sitorus, Syafrudin Abdie
A speech recognition system is an application to recognize spoken words through audio inputs. On the other hand, spectrograms are visual representations of audio signals and effectively represent words numerically. In this research, we developed a spectrogram-based speech recognition system using machine learning's K-Nearest Neighbors (KNN) method to recognize aerospace terminologies, some uncommon for ordinary people. We used 300 audio signal data consisting of two categories, namely 15 spectrogram data for aerospace terminologies and 15 spectrogram data for non-aerospace ones, each taken ten times. By conducting a 70:30 training and testing scheme, strengthened by k-fold cross-validation, the optimum K for KNN is 13. With such value of K, our developed Intelligent system for aerospace terminologies recognition based on spectrogram can achieve an accuracy of 75.56% with a precision of 75.56%, recall of 75.56%, and F-1 score of 75.56%. With that accuracy, our developed intelligent system is considered as a good system. © 2023 IEEE.
Adisutjipto Institute of Aerospace Technology, Aerospace Intelligent System and Data Engineering Research Group, Yogyakarta, Indonesia; Politeknik Negeri Malang, Cognitive Artificial Intelligence Research Group, Malang, Indonesia; Adisutjipto Institute of Aerospace Technology, Faculty of Industrial Technology, Department of Electrical Engineering, Yogyakarta, Indonesia; Adisutjipto Institute of Aerospace Technology, Faculty of Industrial Technology, Aerospace Avionics System Research Group, Department of Electrical Engineering, Yogyakarta, Indonesia