Assessing the impact of (Nigerian, Ghanaian, Zimbabwean, Rwandan and Kenyan) accents on English-speaking speech recognition systems

Date
2019
Authors
Tawo, Ekab-Osowo Samuel
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The use of electronic devices such as computers and mobile phones have rapidly increased over time. Technological powerhouses have inculcated some form of speech recognition in their devices and as such, the increase in usage of this devices provides the increase in interaction between humans and speech recognition. Speech recognition involves adequately transcribing spoken sentences or words into text. This interaction is present all over the world, Africa inclusively, and the continent with its diverse languages comes with its diverse accents and pronunciation of words. A successful interaction between humans and speech recognition involves these systems adequately transcribing the spoken word to the correct text. However, with these transcriptions, errors are prone to occur due to the fact that, mismatch in accents can increase the error rate of a speech recognition system by more than 100% [16]. This Thesis explores to a very large extent the impact that certain African accents such as Nigerian, Ghanaian, Zimbabwean, Rwandan and Kenyan have on the accuracy and transcriptions made by English-speaking speech recognition systems. The findings made in this study states that to a very large extent, the various African accents used in the study does affect the accuracy of the speech recognition systems used for the study.
Description
Undergraduate thesis submitted to the Department of Computer Science, Ashesi University, in partial fulfillment of Bachelor of Science degree in Management Information Systems, April 2019
Keywords
speech recognition, African accents, gender, IBM Watson, Google Cloud, Word Error Rate (WER), transription
Citation