AI cloned my voice
It sounds like me
Hey friends đ I moved this newsletter over to Substack from another platform. If youâre receiving this in your Promotions tab, make sure to drag it into your inbox so that you donât miss any Daily Build updates.
AI for voice cloning
I cloned my voice. This is powerful because I can use my cloned voice model to read snippets of text in other applications and media.
Where this could be useful
Podcast editing - when someone misspeaks you can train the model on that voice and have it say the âcorrectâ phrase or word in post-production
Voiceovers - for blog posts and other digital content (how Iâm going to use it)
Apps and media - generate unique voices for an application or video game
To clone your own voice, youâll need to head over to Eleven Labs. Youâll also need an account which only costs $1/month. They recommend at least a 30 second clip of yourself speaking. For brevity, hereâs a quick clip on what I sound like for reference:
Normal speaking voice
Now after this, I read 30 seconds of text so that the model could train on it. From there, you simply upload the mp3 file(s) and it gets trained on your voice. Roughly 30-60 seconds later you have a voice model.
Then you can just copy/paste some text and upload it to the platform for it to be read by your new voice clone. The following is my cloned voice reading a piece of text on artificial intelligence from Wikipedia:
My cloned voice
This output is pretty good considering the model was trained on only 30 seconds of my voice. There are some parameters to fine-tune your voice including, stability, clarity & similarity enhancement, and style exaggeration. You tweak these variables depending on your desired outputs.
For a near identical representation of your voice, Eleven Labs recommends that you train it on three hours of voice data, so thatâs what Iâm going to do. Then I'll feed my voice model blog posts from Joshuaâs Newsletter to create audio versions for all of my written posts. That would literally save me dozens of hours of work.
Perhaps in the future weâll all have digital AI avatars acting on our behalf, and the content they create will use AI voice models that have been trained on us.
Going next level: Premium subscribers can watch me build something using the Eleven Labs API (an end-to-end product from scratch). You can sign up here.
Daily Build is a newsletter where I publicly explore AI and share my learnings daily.
Get them in your inbox daily: