The design and implementation of a Luganda weather application with speech capability

Abstract

There have been a number of projects undertaken over the years to build weather applications with a capability of emulating close-to-human and almost natural speech for interaction with users. Large tech companies like Microsoft and Apple have over the years have worked continuously on applications with speech capability and have been able to create personal assistants like Cortana and Siri, respectively. Our final year project thus sought to build a weather application, with real-time forecast capability, with a capability for speech, all in Luganda. The Luganda voice from a project of the previous year were not natural sounding and therefore as part of the methodology for the project, we worked to clean speech-corpus based audio data available from recordings for last year’s project done by Mr. Uwizera Methodius and Ms. Katwesigye Phoebe titled “The Design and Implementation of a Talking Luganda Calendar”. We cleaned the audio data using Audacity. We then worked on building a weather application with three modules for different functionalities and a GUI, all in Luganda. Based on the cleaned audio data, we then had to build a clearer Luganda voice using MaryTTS, a text-to-speech platform. The new voice then had to be used within the application to give updates in Luganda.