“15.ai”, 2020-03-06 (; backlinks; similar):
[WP; NN TTS service demonstrating results from custom DL research project by 15 for generating natural high-quality voices of characters with minimal data/few-shot learning; available voices include GLaDOS from Portal and especially high-quality My Little Pony: Friendship Is Magic voices (currently: Fluttershy & Twilight Sparkle); demos: 1/2.
The MLP:FiM voices are trained on a large dataset constructed by the 4chan crowdsourced project “Pony Preservation Project”, begun ~2019. PPP has crowdsourced parsed audio and hand-written transcriptions of all dialogue for all character from all 9 MLP:FiM seasons, the movie, the spinoffs, and various other things voiced by the same voice actresses in case that might help, while processing to remove noise or using ‘leaked’ original data from Hasbro for higher quality still.]
This is a text-to-speech tool that you can use to generate 44.1 kHz voices of various characters. The voices are generated in real time using multiple audio synthesis algorithms and customized deep neural networks trained on very little available data (30–120 minutes of clean dialogue for each character). This project demonstrates a substantial reduction in the amount of audio required to realistically clone voices while retaining their affective prosodies.
I plan to keep this tool up gratis and ad-free indefinitely. This website is intended for strictly non-commercial use.
Thanks to the MIT Computer Science & Artificial Intelligence Laboratory (CSAIL) for providing the initial funding that kickstarted this project two years ago. Further thanks to the Julia Lab, Lincoln Lab, and the Media Lab.
Special shoutouts go to 4chan’s
/mlp/and the anons who have collectively spent hundreds of hours collecting, cleaning, and organizing clips of dialogue taken from the show My Little Pony: Friendship Is Magic. Honorable mention to/g/for some entertaining speculations.And of course, nothing but the utmost respect to the voice actors who originally voiced the characters.