Common Voice for Everyone: Enabling Voice Data Donation in Low Internet Contexts - Mozilla advances platform inclusion with NVIDIA
By Mozilla | Sept. 14, 2022
Voice-enabled functionalities are growing in popularity, yet many people are unable to access or fully utilize this feature. Language barriers and low access to affordable internet continue to lock out marginalized communities from using such technologies.
Mozilla’s open-source voice dataset, Common Voice, is diversifying language use in tech by prioritizing local and low resourced languages. By inviting individuals to donate their voices, communities have a hand in making voice-enabled applications more inclusive and accessible in local languages and variants. But to participate at present, Common Voice requires a strong and stable internet connection, which is a limiting factor. The major contribution components of the platform are the sentence collector and the speak and listen interfaces — which allow users to donate their voice and have their clips validated.
To ensure everyone can easily participate in this initiative, Common Voice invited voice technology enthusiasts last October to provide solutions that would enable users to donate their voices in low and no internet connectivity environments. This initiative was made possible by our collaboration with NVIDIA.
Says EM Lewis-Jong, Common Voice Lead, “By having an organization like NVIDIA - who actually uses Common Voice to build exciting products - invest back into the project, we have been able to advance the platform, address real inclusion issues and ensure that benefit flows back to the open source community. These two winning projects are just the beginning.”
“Spoken voice is the foundation of communication and building speech-based interfaces is one of the most complex machine learning challenges. At NVIDIA, we’re developing groundbreaking research and products in the speech AI domain, and we’re supporting Mozilla as part of a global community effort to extend speech AI beyond the handful of languages currently represented.”
Kari Briski, Vice President Speech AI, NVIDIA
Two teams, one in Ethiopia and the other in Italy, have each been awarded USD $5000, to execute or advance their ideas.
EM continues, “Common Voice is a platform powered by the people and for the people. This open call was to allow communities to innovatively design solutions that will further the accessibility of voice technology. The two winning projects met all our criteria, from accessibility to open-source design, to privacy, and beyond.”
Below, in their own words, the two teams talk about their projects, motivation to participate, and the impact they intend to make.
Limon Data and Analytics | Ethiopia
Our idea was motivated by a previous project that used a Telegram bot to interact with users and provide basic services in low internet connectivity conditions. Telegram is widely used in Ethiopia so we wanted to tap into the existing user base. We hope that this project motivates and enables Ethiopians to donate their voice in the various spoken languages. We also hope that the project inspires other developers and tech enthusiasts to contribute and use Common Voice’s data set to design voice interaction models for African languages. We believe this will make technology more accessible, contribute to de-biasing AI, and overall reduce the technology gap.
Our team is comprised of:
Amanuel Samuel Getahun: A self-taught data analyst and software engineer, involved in projects that challenge existing data related services in Ethiopia. At Limon Data, Amanuel collaborates with governmental and financial organizations to introduce and rebuild data management and analysis structures. He is also a trained architect from Addis Ababa Science and Technology University and a journalist who has worked in various multidisciplinary fields.
Mohammed Abdulkadir Mohammed: A data engineer and software developer who has been involved in Limon Data and Analytics since its conception. Mohammed graduated as an electrical engineer in industrial control from Addis Ababa University and his skills span electrical expertise to software development and machine learning. He leads the operations and development work at Limon where he focuses on tackling the lack of data infrastructure in Ethiopia.
Ruth Ghidey Gebremedhin: An electrical engineering Ph.D. student at New York University where she is studying wireless communications. She has interests in information theory, digital communications, machine learning, data fairness, and haptics. At Limon Data and Analytics, she has formulated different strategies at several levels and worked on designing and enabling different algorithmic functionalities.
Saverio Morelli | Italy
I am Saverio, aka Sav or Sav22999 on the internet. I am 23 years old and a Computer Science student at the University of Trento in Italy. I’ve always been passionate about technology and Mozilla made it easier to build products that made an impact. So I brought this interest home and became a Common Voice community builder in Italy.
This particular project was a canvas to stretch my creativity and contribute meaningfully to growing Common Voice’s open source dataset. I began to develop an open source Android app "CV Project" some time ago, and now I am eager to use this prize money to improve the app’s offline mode features, and its accessibility — and encourage other developers to build similar open source products.
