[Stavros Korokithakis] finds the experience of falling asleep to fairy tales soothing, and this has resulted in a fascinating project that fulfills this desire by using machine learning to generate slightly disjointed fairy tales and read aloud. The result is a fantastic kind of automated, machine-generated audible sleep aid. Even the logo is machine generated!
The Deep Dreams Podcast is completely machine generated, including the logo.
The project uses the natural language generation capabilities of OpenAI’s GPT-3 to create fairytale content that is just coherent enough to sound natural, but not quite coherent enough to make a sensible storyline. The quasi-lucid, dreamy result is perfect for enticing listeners to imagine pleasant nonsense (thanks to Nathan W Pyle for that term) while they fall asleep.
We especially liked reading about the methods and challenges [Stavros] encountered while making this project. For example, he tells us that a great-sounding narration involves more than just pointing a text-to-speech engine at a wall of text and stomping “GO.” A good episode has things like strategic breaks, background music, and audio fades. That’s where pydub – a Python library for manipulating audio – came in handy. As for the speech, the quality of text-to-speech is better than it was a few years ago (and certainly goes beyond machine-generated speech in the 80s), but it still took some effort to find a voice that did it. best at the content, and the project saw gradual improvement.
Deep Dreams Podcast has a GitLab repository if you want to see the code that powers it all, and you can head over to the podcast itself to listen to it.
This post AI-generated sleep podcast urges you to introduce you pleasant nonsense
was original published at “https://hackaday.com/2022/03/26/ai-generated-sleep-podcast-urges-you-to-imagine-pleasant-nonsense/”
