Show HN: AI Shakespeare These days thinking of ways to uniquely show off a text to speech model is tough, as there are a lot of them out there. We thought it would be fun for April 1 to show something a bit on the silly end of things -- we trained our TTS model on "Shakespearean" theatrical voices, and used GPT to generate short plays which are then "acted" out. This lets us show the versatility of our model, that we use in more "serious" settings for generating branded voices for particular languages or accents. It should be noted, the dataset is of our own curation; all voice actors were compensated with full knowledge of being used for TTS and were thrilled to be part of a creative project like this. We resample the speaker embedding space to find voices that are different from any one particular actor. https://ift.tt/Jk53IEi April 1, 2023 at 01:30AM
Comments
Post a Comment