This article was re-written to be more classroom and interactively friendly. A more “serious” and classroom content building article is available here: The Ethical Use of AI in the Classroom.
Let’s cover the basics for the “Too Long, Didn’t Read” crowd. This is not a guide on how to manipulate and blackmail people, or using AI for evil purposes. Here’s the main idea behind this, and it starts with a lovely story from the ancient times – the 1960’s. Back in those ancient, no-TikTok, or even no-Internet years, there was a TV show called Star Trek: The Original Series. On this lovely show there’s a computer that talks back to the people that interact with it – this voice belongs to Majel Barrett. She was the main voice for these computers across multiple series, spanning 6 total Star Trek series all the way up to Enterprise, before she passed away in December 2008. Majel lived a full life, and one of the many things she did was preemptively record audio lines for future uses, which audio engineers can use to create and keep her voice the main computer voice for future Star Trek series and movies. This storage of audio files can be passed upon, processed and create new content for her future generations to see and appreciate her work, but what if a different voice actor passed suddenly in a car accident or something? What if Seth MacFarlane died? Wait, you don’t know who that is? Break time!
Enter AI, because while licensing of the voices is a totally separate thing we’re not going to handle here, we care about new episodes of Family Guy and what not, right? There’s enough content online right now, that we can train an AI with all of it, even after Seth passed away, but what if we couldn’t? What would it take in today’s age to get our wish? About an hour to collect the data, and a few weeks to train the data on a very high end, expensive computer, or purchased online for services that charge $5/hour to work, for 3 to 4 weeks, so it’s not very equitable.
Enter Video Games, because they’re the most important thing in life. Fallout is one of my favorite video game series – and more closer to my heart is a “DLC” game called Fallout New Vegas. If you’re watching the Amazon TV series, they hinted at Season 2 introducing it into the cannon story-line of the Fallout Universe.. and trust me.. there’s a massive one as big as Dune and Star Wars. Robert House, a character in the game more commonly known as Mr House, plays one of the main villains of the game. Man, typing that sentence I just pissed off so many Fallout nerds lol. Well, his name is very similar to a TV show called House, staring Hugh Laurie as Dr. House. A developer already made a simple mod that replaces the model of Mr. House with a Dr. House, but he still sounded like the wonderful voice actor who voiced Mr House, René Auberjonois. I wanted to change that with my mod, Dr. Vegas.
OK, pay attention my non-technical readers, worried teachers and non-video game nerds, here’s the important data you care about; the “Fox News” section is next. No, your students or co-workers, cannot build a custom AI model of you, without your obvious knowledge. Even with hidden cameras and secret microphones, cell phones in pockets, even Bluetooth mics all over the place, it’s not going to be good enough of an audio source. You need direct, silenced, and professionally recorded audio to work with, or else it will NOT be believable in any way. Using the publicly available videos from the House YouTube channel, I pulled as much of their clips as I could using a simple yt-dlp script, and it gave me about 10 hours of audio data to work with. Over the course of the last year, I have been developing and working on a stand-alone, portable toolkit to help me isolate this data of just Dr. House. You can get the toolkit, now fully functional, for free from my GitHub here.
After I used my toolkit, which automatically isolated ONLY Dr. House from all the other speakers in the show, I processed it through a program to isolate only vocally spoken sections. This clipped and trimmed audio now represented about 3 hours of zero sentence-structure and no pauses or breath spaces, of Dr. House saying all sorts of things that did not make any context or reasonable sense. It sounds like “LUPUShouseguitarpianovicodinpseudowilsonchicken”. At time of writing this, I had deleted what’s known as my “training dataset” and cannot post an example of this, sadly.
What we can’t understand, computers can just fine. Re-enter.. or exit and return, to the AI. Using the RVC toolkit, also available free, running on an extremely high end computer of mine ($4500 initial cost, plus about $6000 in custom manufactured equipment), I left it analyzing and creating an AI model, of the Dr. House data, for 3 weeks, 24 hours a day, processing at 100% power. Meaning I couldn’t use the computer to type a document, let alone do anything else on it. I turned a nearly $12,000 AI processing rig, into a school computer for 3 weeks 😉
Here’s the end result of my hard work, Electrical Engineering degree, and sadly, a mental health crisis that left me with some spare development time to get things working fully. Just a side note, you’re not alone, and please reach out to anyone you trust if you’re struggling. The Kids Help Phone is available 24/7, via cell phone, land line, or computer VoIP system, 1-800-668-6868.
Below are three audio files. In order, they are the original line from the video game, with the middle one being the ‘halfway’ process of conversion. You can hear that it essentially removed the accent of Rene, and then made it more sound “flat”. Finally, it put it’s generated audio of Dr House over top, which is of course, the final file at the bottom.
What does all this mean? It means now if someone purchased the license to Dr. House and wanted to create a video game, Hugh Laurie no longer needs to voice it, so he can do other projects and collect the VA credit for his likeness by the pool – and developers can get the AI to record the lines at 3am if they wanted too – no restrictions. THIS is what AI is going to be the future of. Everyone gets to work and relax at their own pace, while the computers do the hard work for us – it’s not going to take your job completely away from you, and if your job is somehow being replaced, then don’t think of it as a down fall, learn how to use the AI and make your job better and more unique. Ask it questions, make it do the hard work. Your kids will have to deal with the robot uprising, not you 😉