| dc.description.abstract | Software programs with Text-To-Speech capabilities (verbal greeting) have been around for quite a while in languages like English, French and German. Our project sought to replicate these systems in Luganda and the problem of the “Lack of localization software programs for native Luganda speaking people, especially the semi-literate and visually impaired, with specific focus on keeping them updated with weather, date and time in their native language.”
In the methodology, we used expert consultation and surveys techniques of data collection to
design the Luganda Greeter and weather updates in a web interface, and then used MaryTTS, a free text-to speech development platform, to build and integrate a Luganda voice in our interface.
Luganda recordings about 2000 were recorded from a radio station studio that were individually split using a software tool known as Audacity. This required phonetic transcription with Java and a data set containing audios previously recorded from an experienced voice talent, their corresponding text and phonetic transcriptions and tags. Finally, we trained the voice using Mary Voice import tools and uploaded the voice to be accessed at the click of a corresponding button from our interface.
Our findings were that previous work in the field of speech synthesis in Luganda was not natural sounding and had poor syllabification and phonetic transcription.
we expect our project to make improvements in terms of intelligibility and naturalness. | en_US |
