Making online educational content accessible through the reformulation of such content in local accents.
It has become increasingly desirable to learn via the internet in developing countries. Most students seeking to learn online would naturally flock towards Massive Open Online Courses (MOOC). These are largely offered in English, a norm reflected in the use of English in scientificc literature.
While there may be positive effects of listening to a foreign accent on listening comprehension in specific younger age groups , this is not more generally the case. Limited prior exposure to a given accent often results in reduced listening comprehension.
This challenge naturally reoccurs in the domain of internet-based learning, where listening comprehension is inherently tied to the utility of a spoken lecture or course. Our project aims to provide greater variety of options for internet-based education by creating algorithmic solutions that allow students to learn online in accents that are more familiar to them, if they so desire.
Same content, different accent. We expect we can achieve this by engineering a generative model which can be used to convert audio between various accents. We conceptualise this as a challenge in accent transfer.
- Formalise the problem of swapping various standard English accents with particular accents local to developing countries, using accent transfer
- Curate a dataset that is consistent with other publicly available audio datasets, which contains mirror counterparts of standard English accents in said local accents
- Demonstrate a methodology by which short phrases can be converted from standard English accents to local accents, and extend this methodology to handle longer audio frames
We find agreement with our motivations for this project in the 4th and 10th Sustainable Development Goals (SGD). The 4th SGD concerns Quality Education, while the 10th SGD concerns Reducing Inequality. If students do indeed choose to use our tool, they will have enhanced access to educational materials, due to better comprehension, which improves the quality of their education. Improved educational quality in turn contributes to the more understated aim of improved equality of opportunity, which aids in reducing inequality of outcome.
Tejumade Afonja is a Graduate Student at Saarland University studying Computer Science. Previously, she worked as an AI Software Engineer at InstaDeep Nigeria. She holds a B.Tech in Mechanical Engineering from Ladoke Akintola University of Technology (2015) and worked on the Fabrication and Design of Robot Vacuum Cleaner for her under-graduate thesis which was published in Alexandria Engineering Journal hosted by Elsevier (2018). She’s currently a remote research intern at Vector Institute where she is conducting research in the areas of privacy, security and machine learning under the supervision of Prof. Nicolas Papernot from the University of Toronto. Tejumade is the co-founder of AI Saturdays Lagos, an AI community in Lagos, Nigeria focused on conducting research and teaching machine learning related subjects to Nigerian youths. She is also an Intel Software Innovator for Machine Learning in Nigeria and 2020 Google EMEA Women Techmakers Scholar.
Munachiso Nwadike: Munachiso is researcher with the Clinical Artificial Intelligence Lab at New York University, Abu Dhabi. He was trained in Computer Science and Mathematics at New York University for his undergraduate degree. While at NYU, his undergraduate thesis was on Semantic Segmentation of Satellite Images, and he built many interesting projects such as a mobile application that interprets sign language with just a smartphone camera. His current work focuses on robustness of deep learning disease classifiers for chest x-rays. Munachiso will be beginning his Masters degree in January 2021 at the Mohammed Bin Zayed University of Artificial Intelligence.
Olumide Okubadejo is a research scientist at Spotify, Paris. His research is centered on automatic and conditional generative music. He holds a B.Eng in Electrical and computer engineering from FUT, Minna, an MSc with Distinction in Artificial intelligence from University of Southampton, and a PhD from Universite Grenoble Alpes. He has authored and co-authored several papers and was a visiting researcher to GeorgiaTech, Atlanta. He is also the recipient of several awards and grants including the Northumbria grant and GEORAMP grant for two years consecutively.
Clinton Mbataku: Clinton is a clinical laboratory scientist, interested in solving Africa’s wide range of health problems using technology. His current research is focused on disease diagnosis using natural language processing algorithms. He is an assistant volunteer tutor with AI Saturdays Lagos where he teaches data science and machine learning.
Lawrence Francis: Lawrence is a programmer with a zeal to understand and build software solutions to problems. He is currently a machine learning research engineer at InstaDeep. His current research is focused on improving sample eciency and generalization of reinforcement learning algorithms and also on the robustness of visual recognition models. He is also a co-organiser at AI Saturdays Lagos where he enjoys understanding, implementing and clearly explaining AI algorithms and was the lead instructor for the Deep learning and Computer vision tracks.
Oluwafemi Azeez: Femi currently works as a research Engineer at Instadeep with focus on reinforcement learning projects. He is a recent masters graduate of Carnegie Mellon University, He spent some time at the African campus in Kigali and in the Pittsburgh Campus where He studied Electrical and Computer Engineering. His research focused on unsupervised domain adaptation in image segmentation, which He did with Yang Zou under the supervision of VijayaKumar Bhagavatula, and speech separation with Yuichiro Koyama under the supervision of Bhiksha Raj. He also co founded AI Saturdays Lagos and Kigali with the focus of helping others learn AI through community study groups,
free online resources and peer motivation.