Statistical Text Analysis for Yorùbá Speech Generation Using Zipf’s Law
Keywords:
Text corpora, Data analysis, Standard Yoruba, Text-to-speech synthesisAbstract
The practical challenge of creating a Yorùbá text-to-speech synthesis has initiated our work on statistical text analysis. Language and speech technology applications have gained an increasingly wide-spread use in several languages/countries, and this has necessitated the importance of examining how much difference exists between English (in most cases the first language for most technologies and applications) and tone languages, specifically, Yoruba. These differences are studied and described in detail in linguistics but they rarely quantified and used by technology developers. In this paper, Yoruba language was described using text corpora from textbooks and newspapers. Other texts from Internet sources were also used. The corpus size was 291,392 word forms and the data was analyzed using Zipf. Based on the statistical analysis, it was found that the coverage of corpora by the most frequent words follows a parallel logarithmic rule for all languages in coverage range, known as Zipf’s law in linguistics.
References
Awobùlúyì, O. Èkó Ìsèdá-Òrò Yorùbá . Montem Paperbacks, Akure, Ondo State, 2008.
Black, A.W. and Lenzo, K. A. Building synthetic voices. available from festvox.org/bsv/bsv.pdf. 2007.
Chen, Y. Zipf’s law, 1/f noise, and fractal hierarchy. Chaos, Solitons & Fractals 45.1: 63-73. 2012.
Dutoit, T. An introduction to text-to-speech speech system. isbn 0-7923-7923- 4498-7. Master’s thesis, Kluwer Academic, 1997.
Fágborún, J. G. The Yorùbá Koiné: Its History and Linguistic Innovations, volume 6. LINCOM Europa, München, Newcastle, linguistics edition, 1994.
Hüning, M. Textstat - simple text analysis tool/ concordance software. available from http://neon.niederlandistik.fu-berlin.de/en/textstat/. visited: April, 2014.
Ìyàndá, A. R. Design and Implementation of a Grapheme-to-Phoneme Conversion System for Yorùbá Text-to-Speech Synthesis. PhD thesis, Obafemi Awolowo University, Ile-Ife, Nigeria, 2014.
Lin, R., Ma, Q. D. Y. and Bian, C. Scaling laws in human speech, Decreasing emergence of new words and a generalized model. arXiv preprint arXiv:1412.4846. 2014.
Manning, C. D. and Schutze, H. Foundations of Statistical Natural Language Processing, volume 999. MIT Press, 1999.
Ngugi, K., Okelo-Odongo, W., and Wagacha, P. W. Swahili Text-to-Speech System. African Journal of Science and Technology (AJST) Science and Engineering Series, 6(1):80-89. 2005.
Odéjobí, O. A. A Computational Model of Prosody for Yorùbá Text-to-Speech Synthesis. PhD thesis, Aston University, 2005.
Odéjobí, O. A. A quantitative model of yorùbá speech intonation using stem-ml. INFOCOMP Journal of Computer Science, 6(3):47–55, 2007.
Owólabí, K. Ìjìnlè Ìtúpalè Èdè Yorùbá (1): Fònétíìkì ati Fonólojì. Universal Akada Books Nigeria Limited, 2011.
Riddle, E. and Stahlke, H. Linguistic typology and Sinospheric languages. First Annual Meeting of the Southeast Asian Linguistics Society. 1991.
Sorell, C. Zipf's law and vocabulary. The Encyclopedia of Applied Linguistics. Wiley Online Library 2012