An IOT-based Language Recognition System for Indigenous Languages using Integrated CNN and RNN

Abstract

Automatic Speech Recognition (ASR) aims to establish communication between humans and computers in a more natural way. The main aim of this study is to build hardware-based automatic speech recognition for Indigenous People (IP)'s ancestral dialects, in particular for Manobo, Mandaya, and B'laan using Raspberry Pi. Jasper is an open source toolkit used for creating voice-activated, always-on applications. The researcher recording audio data from research participants, the study's participants will be located in Davao Occidental and Sarangani for B'laan, Agusan Del Sur for Manobo, and Davao Oriental for Mandaya. A functional microphone and raspberry pi boards serve as the experiment's hardware where audi o input is being fine-tuned from a raspberry pi-powered device that records audio in waveform format, which includes Mandaya, Manobo, and Malita words and phrases. The Tensorflow STFT technique will be used to analyze, generate, transform, and characterize audio signals. JiWER plugins for Similarity measures will also be used The WER output is 98.53%, an acceptable percentage for the number of datasets used