Data-driven approaches to language diversity
A panel at NVIDIA's GTC conference in 2023 discussing data-driven approaches to language diversity, as applied to speech technologies.
Speech recognition through a cybernetic lens
A talk by National Library of Australia and ANU School of Cybernetics for National Science Week where Kathy Reid provides a cybernetic history of speech recognition.
Data visualisation of Mozilla Common Voice v9 dataset metadata coverage in Observable
Covers how accents are represented in voice data, such as BCP-47 and ISO-639 standards, and what challenges this presents for machine learning.
Socially responsible representation of accents in voice data: Considerations for practitioners and policymakers
Covers how accents are represented in voice data, such as BCP-47 and ISO-639 standards, and what challenges this presents for machine learning.
Custodians and midwives: the library of the future
Custodians & Midwives: The Library of the Future provides research and analysis on artificial intelligence and emerging technologies at the National Library of Australia.
Ken Behrens: A cybernetic detective story of algorithms, accents and augury
How can cybernetics help us investigate the story behind a Canberra meme?
Communities are systems: invited keynote for linux.conf.au 2022
What can systems thinking teach us about building open source communities?
DeepSpeech in 5 minutes or less
This Lightning Talk from Canberra Python Users Group provided an overview of DeepSpeech speech recognition, which uses a seq2seq algorithm.
DeepSpeech PlayBook
I developed the DeepSpeech PlayBook to demystify the process of training speech recognition models with Mozilla's DeepSpeech engine.
More choice less voice: The rise of voice interfaces and the decline of open source voice
Voice is becoming ubiquitous. But the way voice has scaled means that our choice in voice technologies is constrained. Let's create voice for everyone, everywhere, in every language.