Josh Seltzer
Full-stack / ML developer by day, primatologist by night. I am interested in the intersection of technology and cognition, and how it will transform our minds, languages, and world(s).
Sessions
Lightning Talks are short 5-10 minute sessions presented by community members on a variety of interesting topics.
When transcribing audio that involves low resource languages, switching between languages, or domain-specific terminology, speech-to-text AI models and existing software solutions leave a lot of manual work to the transcriber. Often this requires first determining which words the AI might have gotten wrong, and then finding resources and investigating by hand to determine what the right words could be. Although the context-specific and low-resource nature of this challenge makes a fully automated solution intractable in most cases, there are existing tools which can be used to support a human transcriber to correctly identify words which they might not be familiar with. However, until now these tools are too scattered and niche for most transcribers to make use of with ease.