We are looking at ways to remaster and restore the entire catalog of Madhouse Live podcasts using Machine Learning/AI technology.
We need help in cultivating a complete sample of Carlito’s voice as well as every single co-host’s voice from 2005-2020, organized into small snippets of audio between 3-10 seconds long.
We will need to devise a collaborative site so volunteers can upload samples or at least give us good timestamps from episodes to pull the audio from.
We have experimented with some software already, without training a Carlito-specific model, and the results have been quite good in separating the background noise from calls. We want to go a step further, though, which is why we’re outlining the project here today.
We are aiming to build a trained model off of Carlito’s voice, which can then be used to surgically extract his voice from every show.
Once we have his voice model trained, we will train everyone else’s voice model, so that we can extract each person from the show audio as a separate track.
Milkman and myself (Akspa) anticipate that this will be a huge undertaking. We want to honor Carlito in the best way possible, using bleeding-edge technology to give listeners a better listening experience.
The end result we are going for is a multi-track version of every episode, which can then be manipulated further with audio editing software to build a fully restored version of the catalog of shows.
Join our discord and enter the #madhouse-restoration channel for more technical details, and to see the latest info/updates on the project.