As part of the ANU’s School of Cybernetics collaboration with the National Library of Australia for National Science week, I gave a talk on Speech Recognition Through A Cybernetic Lens. The talk covered the history of speech recognition, including early attempts such as the Shoebox and the Audrey, and weaved a tale of how competing tradeoffs had influenced the evolution of the technology. In the talk, I also foregrounded the current challenges in the field, including how speech recognition often works poorly for women, people who speak with accents, and other minorities. To conclude the talk, I outlined mechanisms for addressing these challenges, such as gathering more voice data from more diverse and disparate cohorts of people, ensuring that data is labelled correctly, and having mechanisms available to better evaluate speech recognition models.
For one of the slides, I did a data visualisation using data from ABS 2021 Census, and you can see the Observable workbook here:
https://observablehq.com/@kathyreid/languages-spoken-at-home-in-australia-based-on-the-2021-cens
The talk was recorded and is available on YouTube.
To cite this work, you can use the following BibTex:
@misc{reidSpeechRecognitionCybernetic2022,
title = {{{Speech Recognition}} through a {{Cybernetic Lens}}},
year = {2022},
month = aug,
publisher = {{National Library of Australia in conjunction with the ANU School of Cybernetics for National Science Week}},
address = {{Canberra, Australia}},
collaborator = {Reid, Kathy},
langid = {english},
url = {https://www.youtube.com/watch?v=5ROUYa8f4ik&t=6s}
}
