Speech recognition and processing tools are everywhere – from speech-to-text transcription, to voice commands on your phone, to filtering background noise in video calls. However, many automatic speech recognition systems do not perform as well with diverse speakers, i.e. people with an accent, with a speech impediment, or even young or old speakers. Adjusting this technology to recognize different ways of speaking has the potential to make these systems and the tools that depend on them more accessible and inclusive. What challenges and opportunities are there for developing and using more inclusive speech recognition and processing?

In this episode of our podcast series Inclusive Research and Innovation (IRI) in Science, Technology, Engineering and Mathematics, Dr. Claudia Werker from Delft University of Technology talks to her colleague Dr. Odette Scharenborg from the same university. Dr. Scharenborg’s research looks at how to improve speech recognition systems for non-standard speakers. She explains that while acquiring and processing the varied data she needs for her work is challenging, it has also led to opportunities to connect with diverse speakers, to advocate for them, and to help them practically navigate using speech recognition tools.

Learn more about Dr. Scharenborg's research:

• Feng, S., Halpern, B., Kudina, O., & Scharenborg, O. Towards inclusive automatic speech recognition (2024)
• Feng, S., Kudina, O., Halpern, B., & Scharenborg, O. Quantifying bias in automatic speech recognition (2021)