Design and implementation of a Kiswahili mobile learning application incorporating machine learning through audio signal processing approach
Abstract
This project highlighted one of the most challenges underlying the low resource languages. As high resource languages have vast amounts of data for training, Low resource languages have less data for training. Quest for audio dataset for natural language (NL) processing tasks continue to draw research interests globally. Thus a need to target low resource languages ;The basics and basis upon which our project is determined with variable Audio analysis and processing which by farthest remains an important aspect for information processing by computers.
To accomplish the project a visual and descriptive methodology was underlined: this consisted of Research: through reading textbooks, journals, review articles, conference papers, tutorials And Other relevant information from reliable sources, interviews with lecturers and professionals Hardware; laptop and a mobile phone Software: Python programming language, Kotlin programming language, Jetpack Compose, Google Colab, Matplotlib, Ipython, Numpy, Pandas, SkLearn, Keras and other useful packages.
These were put together to have a fully functional mobile application. In this proposed work, a system is developed and demonstrated. The model takes a sound wave as an input and gives a certain metric of similarity as the output. The concepts of "Speech Recognition" and "Pattern Matching" are used to create a Pronunciation Matching tool. This system can be used to enhance the pronunciation skills of the Kiswahili language for people having Kiswahili as their second language. The tool matches the similarity of utterance of a word by a speaker to the ideal pronunciation and gives a percentage similarity or a metric to judge the pronunciation similarity.