We are delighted to share the findings of Biases in Automatic Speech Recognition: Accessibility and Inclusion Barriers in Speech-to-Text Systems, the MA thesis defended by Arunachala Luca de Tena.
Inspired by the UAB use case in ALFIE, this research explores how speech-to-text systems can create accessibility barriers that affect the digital inclusion of certain user groups. The study is framed within the European legal framework and the UN Sustainable Development Cooperation principle of “leaving no one behind.”
Using a qualitative approach, three representative systems (WhatsApp, Apple, and OpenAI) were examined, alongside a literature review of key bias factors such as accent, age, language, background noise, and speech impairment.
The results highlight:
- A lack of transparency in system design,
- A gap between legislation and company practices,
- Random tagging in datasets.
The dissertation underscores the importance of monitoring accessibility and promoting inclusive design that reflects user diversity. By identifying and analysing these barriers, the research opens new paths for building more equitable human–machine voice interaction.
At ALFIE, we are proud to see this work contributing to a more inclusive digital future.