My dad was a gifted preacher. He was wise, creative, passionate, and an excellent storyteller. A few months before he died, before he even knew he was sick, he told me that he had a binder full of his written sermons and other writings that he had been collecting over the years. He wanted to give it to his kids. He said, “I don’t know what you’ll do with it, but it’s yours if you want it.”
A few months later he was diagnosed with lung cancer, and it progressed quickly. Except for the last day-and-a-half when we took him to a hospice house for pain management, he spent his remaining days exactly the way he wanted to: at home, on my parents’ land with my mother, spending time visiting with family and other loved ones.
He died about a week before Thanksgiving, and all of us felt like we wanted more time to prepare for his memorial service. We were too raw and exhausted then to honor his memory the way we wanted. We scheduled his service for Memorial Day.
He said, “I don’t know what you’ll do with it, but it’s yours if you want it.”
Like many people, I have some interest in artificial intelligence programs, and I follow their development with curiosity. I am aware of the dangers they pose and the challenging ethical situations they are creating for us. Like any powerful tool, I know they can be used for harm just as easily as they can help.
That being said, as I looked toward my own part in Dad’s memorial service (creating a video), I became very interested in the idea of using AI to give voice to Dad’s sermons. I was pretty sure that if I could get a decent sample of his voice, I could use a program to clone the audio and bring his written words to life.
The only problem was, I couldn’t find a good enough voice sample. Mom and Dad are/were in their late 70s. They aren’t of the generation that is constantly recording themselves in various formats. I put out multiple requests to family members, and no one could come up with anything usable. Finally my mom remembered that Dad had preached a couple sermons at their local church, and she thought they were on YouTube.
I found them. One was completely unusable. The other gave me only a little hope. I pulled the audio from the sermon and fed it into a voice cloning program (https://elevenlabs.io in case you’re interested). It didn’t work. The sample audio quality was too garbled. Getting desperate with less than a week to go before the memorial service, I hopped onto Fiverr.com and looked for an audio engineer who might be able to clean it up enough to be usable. I eventually found a guy in Pakistan with great reviews who promised to do the best he could.
At first I still wasn’t sure the audio would be usable. It still sounded like a bad recording, just slightly less so. But then I fed samples of that somewhat improved recording into the the eleven labs AI program, and somehow it made it sound a little clearer, stronger.
I became very interested in the idea of using AI to give voice to Dad’s sermons.
The program has settings in which you can describe the person speaking. I described my dad: his age, ethnicity, region, style of speaking. I also described him as having a clear, strong voice, which seemed to help the quality of the audio. There were a handful of other settings to tinker with as well. I spent a lot of time adjusting them to see which ones made it sound more like Dad.
Once I had the voice in good shape, I started feeding the program sections from Dad’s written sermons. Through trial and error, I learned that it sounded good for about 1-2 sentences. After that the cadence or speed would get off, sometimes speeding up or slowing down, emphasizing random words.
It took a lot of time and tinkering to get it to sound like Dad. I’m sure it would have been easier if I had started with high quality source material. (Or, if I started this six months later, since AI programs are becoming more powerful every day.)
The other tough part was that it was emotional work! I would start digging into the technical problem, but the more I improved the audio, the more it sounded like Dad, and I periodically had to stop to cry.
My wife Emmy was helping me the entire time. She and I both searched through the binder with hundreds of sermons/writings to find ones that could tell Dad’s story in his own words. We had almost no video, so we sifted through photos and tried to find ones that would help people remember his life.
I did find one video clip of him walking away from the camera. I had taken it several years ago on a brothers/dad weekend in the early days when I was learning/practicing how to make videos. That shot of him walking away at the end of the video still makes me tear up almost every time I watch it. I’ve heard the same from many others.
Dad was an emotional person, and that carried through into his preaching. The AI program couldn’t entirely capture that, but adding a soundtrack helped.
The more I improved the audio, the more it sounded like Dad, and I periodically had to stop to cry.
The end result, while not perfect, felt pretty good to me. When it was played at the memorial service, no one (except for the few people who already knew) suspected that it was an artificial voice. Everyone, people who knew my Dad his entire life, assumed I had somehow acquired recordings of his sermons from past years. When I described the process of making the video, many people reacted in a way similar to my uncle, who said something like, “Boy, I don’t know. That video was beautiful. It was perfect for today. But it does make me worry about all the other things AI can do.”
I don’t know that I have the energy for—nor interest in—trying to imagine all the negative or harmful ways AI can and will be used. This isn’t really a post about the ethics of AI. It’s a story about how I used a new tool to try to make something meaningful and beautiful. Humans are not going to stop making powerful tools. The real question is, how can we use those means to make life a little better?
Share this post