NotJordanPeterson.com
I'm a software engineer with background in machine learning, robotics, and game development. I've been working on a side project recently, exploring paths for starting a company to make content creation tools powered by the latest machine learning techniques. I just released a pr
Trending on Indie Hackers
Nice work! I posted on reddit but might as well post here. What's the backend looking like for u? Are you serving on gpu instance? Or are u serving TTS and vocoder seperately?
Also how many of his audiobook did you use? Like how many hours of data? In asking because the model's pronounciation is generally pretty good. But Jordan does have a diverse vocabulary in his books.
BTW, I tested the model on some words from Harvard sentences, it seemed have trouble pronouncing some words like "salmon", probably Because Peterson didn't say those often. I recommend training a model on phonemes instead of graphemes, this way there's less pronounciation error and you can coax the model to say difficult to say words. As currently on top of learning TTS your model is also learning a G2P (grapheme to phoneme)
Awesome work!
Thank you! And thanks for the feedback. I'm serving on a GCP instance with a single K80 right now, which is why the site is running so slow. Didn't expect so much traffic. :) Serving Tacotron and Waveglow on the same GPU. Trained on about 20 hours of data. You're right it does get a lot of pronunciations wrong. Good point about training on phonemes. I want to try that at some point. Was less work to do graphemes to start, and just wanted to get something working. Thanks for checking it out.
This is both hilarious and amazing (❤️JP btw)
hey be
This is super interesting. I'm going to play with this and see how it would work for screencasts. I find someone like binging with babish has a great voice for audio. I wonder if NJP could also create a soothing instructional voice for my project www.uecasts.com
Edit: Got a 524 error when trying it :(
Thanks! There are definitely lots of applications for this technology (speech generation in general, not celebrities in particular). Sorry it didn't work for you. :( I wasn't expecting such high traffic. It's been getting slammed all day. I might have time to add some more resources to it this weekend, but for now what's there will have to do. If you keep trying it should eventually work. The site hasn't crashed yet, it's just been running slow.
Fantastic, in future a directory of voices to “hire”, however i wonder if it’s legally safe to use somebody else voice without permission.
Hahaha, this is wonderful! Awesome work!
How much input would this kind of thing need in order for me to reproduce my own voice with it? e.g. would I be talking into a mic for an hour? 10 hours? 100 hours?
Thanks! This was trained on about 20 hours of data. Haven't yet experimented with how little I could get away with.
Neat! That's a lot but not too bad. Not sure why I'm fascinated with having a bot that speaks like me but it seems fun :P
This comment was deleted 6 years ago.
I wrote a summary of things back when I disabled the site. You can find it here: https://bit.ly/39WWn71. Still working on the tech and a few applications, but that site will remain disabled for now.
This comment was deleted 6 years ago.
This comment was deleted 6 years ago.
This comment was deleted 3 years ago.
Thanks! I just started learning a lot of this stuff a couple months ago as a side project. Most of it has been me learning a bunch of things, but it's been fun. Start coding about 7 weeks ago, but if I had to do another it would probably only take a week or two with the same setup.
This comment was deleted 3 years ago.
This comment was deleted 6 years ago.
Indeed! Looks like I may need a beefier instance. :) It's still running, just a bit slow. I'll have to look into some optimizations. Nice to see that there's interest though!
This comment was deleted 6 years ago.
This comment was deleted 6 years ago.
I also posted on reddit last night. :)
This comment was deleted 6 years ago.
This is definitely where things are headed. In fact the startup I'm working on will focus on voice acting (not celebrities, but synthetic voices) for games/films in part. There are a few companies already working on text to speech for web articles as well. Going to be an interesting couple of years coming up.
This comment was deleted 3 years ago.
This comment was deleted 3 years ago.