DEEP LEARNING ASSISTIVE MODEL IN NIGERIAN MAJOR LANGUAGES FOR THE RURAL COMMUNITIES USING JOINT GMM REGRESSION ALGORITHM

The use of indigenous languages for developing assistive models, particularly in the context of Artificial Intelligence (AI) and Machine Learning (ML), focuses on overcoming digital divides, enhancing accessibility for people with disabilities, and promoting language revitalization. Deep learning models for Nigerian major languages (Hausa, Igbo, Yoruba) are primarily focused on text-to-speech (TTS) synthesis using FastSpeech 2/HiFi-GAN, neural machine translation (NMT) with transformers, and alphabet/speech recognition systems. This work focused on Deep Learning Assistive Model in Nigerian Major Languages for the Rural Communities Using Joint GMM Regression Algorithm. The research conducted in one of Nigeria federal universities revealed that over 70% of the rural communities members cannot read and write in English language. Prior to the advent of Europeans, all the Nigerian ethnic groups had their own distinctive cultures, traditions, languages and indigenous system of education. Their medium of instruction was their tongue. The first formal (western) education was introduced in Nigeria in1843 by Christian missionaries; their aim was to convert the heathen African to Christianity through education. Nigerian formal education was patterned after the English system.  This among other things makes access to Information Technology (IT), Mathematics, Science and Technology a difficult task for this category of people. Deep Learning Assistive Model in Nigerian major languages for the rural communities using joint GMM regression algorithm will go a very long way to bridge this gap. Since we are dealing with three (3) major Nigerian languages, joint Gaussian Mixture Models (GMM) regression algorithms was adopted. The model will promote digital inclusion for the rural communities and individuals living with impairments and bridge language barriers, utilizing datasets from platforms like Mozilla Common Voice. Individuality speaker voice is the result of many acoustic and linguistic cues. Individuality speaker will be described only by segmental acoustic characteristics. In particular, vocal tract parameters and LP residual signal, as speech model parameters, will be converted by the conversion of prosodic features. Source data with its corresponding parallel target data will generate the needed result. State of the art vocal tract conversion systems are based on GMM  to model the speaker acoustic space and map the vocal tract parameters.