TINKERED THINKING IS FAKE

March 29th, 2021

Recently, Tinkered Thinking made a change.  If you only read the platform then you probably haven’t noticed, but if you listen, chances are you’ve picked up on something different.  Yes, Tinkered Thinking has outsourced the recording of podcast episodes.  Sort of.  It’s still the same voice, but it’s no longer me reading episodes.

You may have recently seen videos of Tom Cruise which are not actually videos of Tom Cruise.  These are so called ‘deep fakes’ which use a variety of machine learning techniques to generate hyper-realistic video and audio.  Tinkered Thinking has begun to employ something similar, though not as perfectly effective.

First, a small confession: I, the creator of Tinkered Thinking, absolutely detest recording episodes.  In the beginning it was merely difficult and I figured like most things I would get fairly good at it and find a way to enjoy it as a kind of practice.  As luck would have it, or rather not have it, I haven’t improved much in my ability to read and record.  I still make loads of mistakes that have to be edited out, and the entire process lacks all sense of satisfaction and fulfillment that comes with other things, like writing everyday, for example.  

So, when the discovery came about that it’s possible to deep fake your own voice, and then simply plop a transcript into a program that then uses a machine voice to create pretty decent simulacrum, I couldn’t resist.  This required reading a half hour of Planet Earth script which is then used to train the machine on the sound of my voice.  Considering the cost and the ease and good fidelity of result, I couldn’t pass up on the use of this tool.  Granted, it isn’t perfect.  It’s certainly robotic, and it lacks some nuance about exactly how some things should be intoned.  

If Tinkered Thinking had more audience support then I would consider continuing recording episodes myself, but alas Tinkered Thinking is still just a little hobby project.  One that I certainly want to continue.  And in light of that fact, if you would like to support Tinkered Thinking you can do so on the support page.  In the meantime, while the Tinkered Thinking audience and income is still relatively small, this machine voice offers a substantial improvement for insuring that Tinkered Thinking continues:  it makes the production of the podcast far less time consuming and to be perfectly honest, the whole project is more enjoyable with the removal of it’s most unpleasant thorn.  This is on top of the fact that Tinkered Thinking is just a side project, and my main focus independent of Tinkered Thinking is ramping up into what will probably be a very busy year.  This technology offers an edge against being busy: the chances that I can keep up with the project are higher if I can get it done quicker. 

There’s also the consideration that this deep fake technology which Tinkered Thinking uses will certainly get better.  It’s doubtful that it won’t improve to the point where I wouldn’t even be able to tell the difference between an episode I personally record, and one generated by a more advanced deep fake version of my voice.

The particular technology being used here is a program called Descript, and the feature is called Overdub.  It’s improved quite a bit in just a year, to the point where it makes sense to actually use it, and everything points toward it getting even better.

If you truly can’t stand the change, feel free to reach out on Twitter, and I’ll be happy to discuss it further.  Though, if you aren’t one of the platform’s generous and treasured supporters, I can’t promise your argument will land with as much emphasis as you might hope.  Surely anyone can understand why.

At the very least, though the result might not be perfect, it is pretty cool regardless.  It’s a perennial wish among so many people to have clones of ourselves so we can get more done and enjoy more of the various facets of life.  Well, we now live in a time where we can get a taste of that and outsource some of the less fulfilling work to a digital version of ourselves.  

As a final note, Tinkered Thinking will still continue to record the weekly Lucilius Parables since the machine voice is particularly inept with the vicissitudes of fiction, not to mention that those episodes are actually kind of fun to record.