Research Article
Shared Syllables for Amharic Tigrigna Text to Speech Synthesis
@INPROCEEDINGS{10.1007/978-3-030-93709-6_37, author={Lemlem Hagos and Million Meshesha and Solomon Atnafu and Solomon Teferra}, title={Shared Syllables for Amharic Tigrigna Text to Speech Synthesis}, proceedings={Advances of Science and Technology. 9th EAI International Conference, ICAST 2021, Hybrid Event, Bahir Dar, Ethiopia, August 27--29, 2021, Proceedings, Part I}, proceedings_a={ICAST}, year={2022}, month={1}, keywords={Bilingual text to speech Festival Syllable Amharic-Tigrigna}, doi={10.1007/978-3-030-93709-6_37} }- Lemlem Hagos
Million Meshesha
Solomon Atnafu
Solomon Teferra
Year: 2022
Shared Syllables for Amharic Tigrigna Text to Speech Synthesis
ICAST
Springer
DOI: 10.1007/978-3-030-93709-6_37
Abstract
In this study, an experiment is conducted to explore and exploit shared Amharic and Tigrigna syllables in the development of Amharic Tigrigna bilingual text to speech synthesizer. Both Amharic and Tigrigna are under resourced languages, yet these two languages share the Geez writing system with large portion of phone sets and syllables. This study therefore shows the possibility of constructing Amharic-Tigrigna bilingual text to speech synthesizer based on the shared syllables to optimize linguistic resources. The dataset for training and testing is composed of consonant-vowel syllables in both languages. Festival speech synthesis framework is used for the experiment. The result shows mean opinion score of 3.09 and 2.08 for intelligibility and naturalness, respectively. Epenthesis vowel insertion and possibility geminates which are not predictable from the text at surface level in both languages greatly affect naturalness of the synthetic speech. Another factor that affects the naturalness is the fact that we used an already existing multilingual speech synthesis framework that has foreign accent. Even though the naturalness is below average because of the aforementioned reasons, the possibility of exploiting shared features to develop multilingual speech synthesis for under resourced languages is encouraging. We have learned that to enhance the performance of the bilingual synthesizer, there is a need to integrate language specific features.
