A big new development in AI has sprung up on us. Using ElevenLabs Speech Synthesis we can now easily train a Machine to learn your voice (I did my voice with only three youtube audio clips) and then have it say whatever you want in the style your voice.
Hear my AI-generated voice below (and a Poem Open-GPT was nice enough to craft for me)
The results are easy to produce (signing up for a free account and accessing the 1 Month Free Trial to Train your own Voice) and result in a very human-sounding response that sounds very much like the trained target.
Perhaps I will soon be replaced, or maybe that has happened already…
Interesting! It seems to have captured timbre pretty accurately, but superimposes a bit of an American accent. I wonder if the model is over-fit in that sense!
Do you think if you trained it on a greater sample of videos the American accent would begin to disappear?
If the training data is still being developed you may end up with the AI making everyone else sound Australian if you feed it enough Core Electronics videos
I do think more samples would result in a better accuracy (less American accent). This system currently was trained on only three, less than 10MB large, MP3 Files. It remarkable.
And Streuth, Long may the Australian Accent Reign! We best all get into it.
If you want to get really worried about your job, I asked chat GPT to write a setup shell script for me. It took a few iterations and I did make a few adjustments, but it did a really good job.
I have used Eleven Labs and honestly paying 20+$ for an AI is a bit too much. The AI voice model I’m currently using has TTS, dubbing, video/audio translation, voice cloning and much more but at a way lower pricing
Well, to be honest, this is a bit expensive price for an AI tool. There are many AI voices available online which are much cheaper then this and almost have the same features like voice cloning, video translation, subtitle generator etc.
Two good points from @Anni255402 and @Khushi259119 - AI has moved pretty fast in 2023 and I’m sure each service that gets released is met with a fair bit of competition. Even the original post which is >10 months old is probably an obsolete offering by now. Thanks for the links folks
Thanks for sharing this exciting development! AI voice synthesis has come a long way, and it’s amazing to see how accessible and realistic these tools are becoming. If you’re experimenting with AI voice technologies, I’d recommend also checking out https://audiomodify.com/.
It’s another fantastic platform for creating and customizing AI-generated voices. They offer a variety of tools that can enhance your experience, and it might provide some alternative features or workflows to complement what you’re already doing with ElevenLabs.
Looking forward to hearing more about your projects!
Totally agree — it’s pretty impressive what the system managed to do with just a few samples. Less than 10MB of data and already such a solid result? That’s no small feat.
And yep, more samples would definitely help fine-tune the accent further. Australian English has such distinct nuances — would be great to preserve that clarity and identity in AI-generated voices too.
