koleshjr/Mozilla-Foundation---Mozilla-Common-Voice-Hackathon-I-Nairobi-

Kiswahili Automatic Speech Recognition (ASR) Model Training # Mozilla-Foundation---Mozilla-Common-Voice-Hackathon-I-Nairobi- Kiswahili Automatic Speech Recognition (ASR) Model Training: Mozilla Common Voice is an open-source initiative to make voice technology more inclusive. Supported by the funding from Bill and Melinda Gates, FCDO and GIZ, we have been enhancing the growth of open source data sets for Africa Languages in East Africa. Our open source data set for Kinyarwanda has been used to build a chatbot to respond to COVID-19 questions, and our partners in Uganda are working on a keyword spotter in Luganda. In 2021, we focused on building an open source data set for Kiswahili and now Mozilla is opening up this competition as part of a series of activities that work towards equipping the technical community in East Africa with skills to build voice technology. This Hackathon, and the wider series of activities, is aimed at increasing the impact of the MCV Kiswahili dataset by fostering a developer community with the skills to take the dataset and make use of it in problem solving. The objective of the challenge is to train an ASR model for Kiswahili using MCV Kiswahili data. Beyond building a well performing model, we are keen on models that have comparable performance across demographic characteristics, particularly age and gender. We will compare the overall performance of the model with how it compares on the following evaluation sets representing the following demographics: Women Women under 30 Women over 30 Men Men under 30 Men over 30 ## How to use this: navigate to src folder download the training files and use GPU in training ## Acknowledgements In most of my research I followed @osinkolu 's wonderful notebooks: github.com