Audio Data for Machine Learning

Hi HN. I'm working on a new project in the realm of data-centric ML. Audio data in machine learning is quite limited, which in turn limits the types of tasks we're able to learn to do (speech to text, text to speech, etc). I think that much more is possible, if only we had the right data.

I'd like to start creating more, varied labeled audio data for ML, and I need the community's help to figure out what is needed. Please reply in this thread, I'd love to hear your thoughts.

Thanks!