ElevenLabs. Lots and lots of ElevenLabs Text to Speech stuff… I started to get interested in AI and Machine Learning with an emphasis on chatbots when I learned about Replika in 2022. My Eventually, I dove in and wrote some Python code that got two replikas to talk to each other by passing what they were saying back and forth between two browser windows.
Then in Fall of 2022 I started messing around with OpenAI GPT-3. The responses were much, much better, even with some humor about the Foo Fighters vs. the Goo Goo Dolls.
….but it was soooo slooooooow… Holy smokes. And this was the state of the art – millions and millions of dollars of hardware serving up Text Generation before ChatGPT became a thing…
Shortly after that, ChatGPT Also, this is when I first started using ElevenLabs for Text to Speech instead of xVASynth. The speech quality for text to speech vaulted through the roof.
It was improving, too much! I had trouble getting short conversational answers out of ChatGPT – however, ElevenLabs Text to Speech allowed me to realize a dream I’d had since probably 2006 when I started playing World of Warcraft. My original WoW character could actually converse with me in her own, original voice “cloned” from the 10 or 15 seconds of /silly audio of jokes from the original files from World of Warcraft!
(albeit in an excessively wordy and very unnatural fashion. But I’ll take it.)
Since then, it’s been one project after another.
I’m using ElevenLabs Text to Speech to generate the Text to Speech for the Race Recap audio for my iRacing League’s races three times a week.
Most recently, I’ve been working with Mantella, a program that brings AI and ML chat to NPCs in Skyrim. Currently the Text to Speech engine is once again limited to xVASynth, but there are new techniques that should allow a substantial increase in quality coming out this month – January 2024.
