Text-to-speech for Ghanaian language (Akuapem Twi) on an embedded system

Abstract

This paper presents the development of an End-to-End Text to Speech (TTS) system for Twi, a language spoken in Ghana. The aim of this work was to provide a solution for synthesizing speech from text in Twi language, which can be beneficial in various applications, such as language learning and accessibility for visually impaired people. Two different versions of the TTS system were implemented, and tests were conducted to evaluate their performance. In addition, we tested the system on an embedded platform, namely Arduino Nano RP2040 Connect and Raspberry Pi, to assess its feasibility for deployment on resource-constrained devices. While one implementation produced static sound, the other successfully generated Twi speech output with satisfactory quality. The performance of the first implementation was evaluated by plotting the training and validation losses against the number of epochs and drawing conclusions based on the results. The remainder of this paper is organized as follows. Section 2 describes the related work and existing TTS systems for Twi language. Section 3 provides an overview of the proposed TTS system, including the architecture, dataset, and training process. Section 4 presents the results of the experiments and their analysis. Section 5 concludes the paper with a discussion of the limitations and future directions for improving the proposed TTS system.