This research seeks to develop a phonetic detection system for the Kambera dialect, the East Sumba local language, based on the TensorFlow framework that will be implemented in mobile applications. As part of this initiative, this research compiled a representative dataset of Kambera dialect phonetic samples. The main objective is to improve precision in phonetic recognition. Using the Kambera dialect as a case study, the data were extracted and trained using the open-source Teachable Machine service. This research adopted a convolutional neural network (CNN)-based approach combined with the Mel-frequency cepstral coefficients (MFCC) method for more accurate feature extraction. After data collection, model training, testing, and implementation, the model was integrated into the Android platform to benefit the public who wished to understand the Kambera dialect of East Sumba. The development and testing of this system were designed to detect and interpret the phonetics of the local language of East Sumba with the Kambera dialect, making a significant contribution to optimizing phonetic recognition and providing a dataset for ongoing research interests. It also serves as an accessible linguistics educational tool and supports linguistic inclusion and diversification in digital technology. Empirical evaluation showed that the overall average dialect detection precision rate reached 98.3% to 99.6%, with the user satisfaction rate reaching 99.33%. These results confirm that the developed system has a very efficient and good detection capability.
Copyrights © 2023