Hi Friends,
I have spent 3 painstaking weeks voice training the Real JARVIS voice from Paul Bettany. I have tested and its spot on. I have the onnx and json files and I am willing to share. My question is, am I allowed to post them on here, just thought i would check. I know there are a lot of people on here that would like these for authenticity, i haven’t found any that come close in my testing so i created them. I Look forward to contributing.
Not sure if you can post them here, but why not share them on Github?
I would love to try it.
Sure, i will try and put them up over there.
Ive been mucking around with the inference length as some of the words seemed a bit rushed someone with a bit more experience should be able to tweak it to perfection.
Update, tried to update them to github but it says 25mb limit. any other suggestions on how i can share? I dont want to violate any terms.
ok figured it out here are the files. Enjoy.
How did you mimic the voice? What was the way to go?
I would love to do something like this for a dutch voice.
Can you implement a media player in the page, so we can hear it before downloading?
i used audacity to capture the voice, then fed it through voicebox and generated the harvard 300 sentences in that voice, then trained it through piper and pytorch. Some words are a bit off, i am not sure how to fix that yet, but someone with more experience might be able to tweak it. The Cadence and pitch is pretty good.
As for a demo, you can just download the files they are small and then create a voice assistant to text them yourself, it only takes a couple of mins.
Mads Mikkelsen. I might start using a voice assistant for that voice 
Thanks for creating the Jarvis real voice. I have modified below given settings to keep the voice in slightly lower speed. YMMV.
- Navigate to Settings
Add-ons Piper. - Click on the Configuration tab at the top.
- Look for the length_scale (or overall speed) parameter.
- Change this value above or below the default 1.0 (I have used 1.1)
- Click Save and Restart the Piper add-on to apply the changes.
With his permission and royalty payments? Some voice actors do this for a living. DMCA takedown notices can be embarrassing and inconvenient, especially if you have spent many hours on your project.
I completely agree with the sentiment regarding respecting terms and copyrights—in fact, I asked in my original post whether hosting this was permissible here.
To clarify, this model is a digitally synthesized approximation built through training data and algorithmic reconstruction rather than an exact, raw audio likeness. While it is certainly close, it is distinct from his actual recorded voice.
That said, if a formal takedown request or notice is issued, I will gladly take the repository down immediately. I appreciate the heads-up and will keep an eye on how similar AI voice projects across platforms are handled moving forward.