Amplify: Empowering Underserved Voices in Spoken Language Interaction
by Simon Robinson, Thomas Reitmaier, Jennifer Pearson
The Amplify project is focused on expanding the reach of AI-driven speech technologies, looking ahead to a future in which they are available for everyone, particularly those whose voices have historically been underserved by mainstream AI systems. By embedding RRI and EDI principles throughout its work, Amplify aims to demonstrate a new vision and blueprint for fairness, accessibility, and inclusivity in spoken language interaction.
Why Amplify?
Of the 7,000+ languages spoken worldwide, only a tiny fraction are supported by existing spoken language systems. In addition, while speech recognition technologies have made incredible advances, many systems continue to struggle with diverse accents, dialects and speech patterns, not to mention the realities of everyday speech, where slang, code (language) switching and diverse ways of speaking can make such systems entirely unfit for purpose. This leads to exclusion, frustration and missed opportunities for language communities who do not fit dominant language models.
Amplify is tackling this challenge head-on by demonstrating that AI-powered speech technologies that are developed in an inclusive manner are not only more suited to the ways their speakers actually talk, but can lead to new ways of interacting with voice content that can better benefit the communities that help create them.
Project Activities
Building on research from the EPSRC-funded project UnMute: Opening Spoken Language Interaction to the Currently Unheard, a key activity during Amplify is to facilitate uptake of the UnMute Speech Technology Toolkit, working with a network of community partners and NGOs to prove and refine the tools and build a network around equitable spoken language technologies.
The UnMute Toolkit was formally launched at the Indian Institute of Technology Guwahati in January 2024, and the project team have since undertaken test-and-challenge activities with language communities in India, South Africa and Kenya. As the project continues, we will update and improve the toolkit, and engage with policymakers, researchers and practitioners to promote and facilitate not just this specific set of tools, but our long-term goals for more inclusive, responsible speech technology development.
Embedding Responsibility
Responsible research and innovation is at the core of the Amplify project. Currently, mainstream innovation in speech technology is focused primarily on the affluent minority. In contrast, our work focuses on laying a foundation for creating speech-driven interactions targeted at the many languages that lack commercial viability, and are therefore typically overlooked (or worse, seen simply as data sources to be exploited rather than partnerships). It is this inequality that makes the project a fundamental part of “levelling the playing field”: to include such languages – and language communities – in a responsible manner, and ensure the benefits speech systems can provide are extended to as many people as possible.
Our co-design approach ensures that the communities that we are working with and who contribute language information will be direct beneficiaries of the new technology that we develop. Fundamentally, our approach is to ensure the people we work with are truly engaged in the innovation process, stakeholders are engaged throughout, and the implications and impacts of our work are thoroughly considered.