About
Revolutionising communication with a convenient and less stigmatising Automatic Speech Recognition (ASR) solution for those with slurred or hard-to-understand speech.
Problem
Automatic Speech Recognition (ASR) is transforming the ability of people with communication difficulties to communicate.
This can now be deployed on mobile devices, making it more convenient and less stigmatising for people with slurred or difficult-to-understand speech.
However, this technology is not available to those who would most benefit - non-English speaking people in African countries, who often have the least access to traditional assistive technology and speech and language therapy.
Solution
The creation of a Centre for Digital Language Inclusion (CDLI), which in the first 18 months will:
- Collect datasets of non-standard speech for 10 African languages Collate the recordings and build ASR models in parallel.
- Develop AI-driven inclusive communication technologies.
- Build a community of innovators and users across Africa to maintain open-source applications.
- Provide training to users and ecosystem enablers (e.g. speech and language therapists) to ensure the technology is adopted. Data sets will be open source.
Outcome
Greater access to and equality of communication the millions of people living with communication impairments in Africa through the development of open-source datasets, ASR models and applications. These will be built for and with the community.
What We Have Done
AT2030 programme, led by the Global Disability Innovation Hub (GDI Hub), has tested freely available AI-enabled apps to help people with non-standard speech to understand how the app is currently being used, how could the app be used in future, and how it could be improved.
Conducted training in Ghana for speech and language therapists to support the use and development of this technology and it's application for those that need it most
Why it matters
- 28-49% of people with disabilities worldwide experience difficulties with communication at some point in their lives.
- In African countries, these barriers are often more pronounced as a result of compounding factors that range from the lack of Speech and Language Therapy (SLT) services and poor availability of assistive technologies to support communication, to stigmatising cultural beliefs.
Objectives
Year 1
- Collect 5 non-standard languages in Ghana
- Develop ASR algorithms
- Trial makeathon for communication applications
- Support 3 ventures to get on app stores
- Collect bulk language cards for Google Relate
- Develop API for Google Relate for bulk cards
- User test Relate with bulk uploads
- First-year report
Quotes
“For the first time, our program offers practical help to vulnerable communities in Africa living with non-standard speech to be better understood in their language, not in the language technology dictates. We support those who are used to living with the stigma of talking differently to have a better opportunity to live the life they wish for and participate in education and work – because they are better understood.”
“We will create a ‘cookbook’ of how to build open-source data collection for non-standard speech, speech recognition models, and tools that enable local innovators to build apps for local languages that will support their communities, wherever they are in the world.”