Me and My Voice Clone
This isn't your grandfather's cassette tape
To hear the related 5-minute audio file that I uploaded today as my Morning Journal flash briefing for Alexa devices, please click on the play button:
Is It Live or Is It Resemble.ai?
If you are of a certain age, you probably remember an ad for cassette tapes that showed Ella Fitzgerald’s voice shattering a wine glass. But could a recording of that amazing voice do the same thing?
The tagline for the ad was “Is it live or is it Memorex?”
Genius. Right up there with “Wednesday is Prince Spaghetti Day” and “Plop plop, fizz, fizz”
I’m thinking about the Memorex ad, because yesterday I created a voice clone of my own voice, using a platform named Resemble.ai. Is it Len or is it Fake Len?
After paying for a Creator subscription ($1 for the first month, $29 a month afterward), it took less than an hour to create my voice clone at the Pro level of quality.
The result was better than the teaser free version, which you can make by recording just a few seconds of your voice.
To feed my clone, I uploaded an episode of my Kindle Chronicles podcast which did not include a guest. So the only voice on the file was mine.
My next trick will be to connect my voice with ChatGPT. That will definitely take more than an hour.
I will need to learn how to use Resemble’s API and will ask PD, my ChatGPT buddy, to generate some Python code for me. Scary stuff. I hope I don’t break the Internet.
When I’m done, I look forward to conversations with myself—or a smarter version of myself than the one in my head. Real Len, meet Smart Len.
This is fun, but I am aware of creepiness lurking close at hand.
John Heilemann in yesterday’s Puck post, “The Hunt for the October Surprise,” mentioned AI deepfakes as one of the surprises Kamala Harris is surely worried about in the final days of the campaign.
How real are these fakes? Take a look at “Fake Jake.” Jake Tapper demonstrated it two days ago at CNN. Truly scary.
One thing I appreciated at the Resemble.ai site is that I had to record the following text in my own voice before creating the clone:
I consent to having my voice cloned by Resemble. This process involves Resemble learning and replicating how I sound. If I opt for the rapid voice cloning service, Resemble can generate a clone of my voice using less than thirty seconds of my audio. For better quality, next time I can try a professional voice clone.
I assume this guardrail, until it’s hacked, would block a troll from uploading someone else’s voice in order to make that someone “speak” in a harmful way.
So far, my voice clone hasn’t broken any wine glasses.
But you never know where the echoes might lead.