Tagalog regional accent classification in the Philippines

Abstract

Accent classification has been a focus on recent computational researches since it directly influence the performance of automatic speech recognition technologies. In this paper, we aimed to automatically classify Tagalog accented speech of speakers from Region IV-A, Philippines. Speech and voice data were collected from 150 local residents with strong accent from the 15 towns of five (5) provinces of the region, namely: Batangas, Cavite, Laguna, Quezon and Rizal. The data gathered was cleaned and denoised using Audacity sound editor software. We then extracted some voice features from the cleaned data using PRAAT application software. These include: harmony, pitch, intensity, power, LFCC and MFCC. We tried several data mining tools to address our objectives. Results showed that MultiLayerPerceptron (MLP) classifier gave the most significant result. Among the towns that have distinct variety of accent are: Talisay, Batangas; Maragondon, Cavite; Paete, Laguna; Lucban, Quezon; and Taytay, Rizal. The significant features that classifies tagalog accent among these towns are: standardDeviationPitch, maximumHarmony, minimumIntensity, standardDeviationIntensity, minimumLPC, meanLPC, LFCC and standardDeviationMelFilter.