BrainDamage, a neural network synthesizer - feedback welcome
- KVRist
- 46 posts since 2 Jul, 2021 from Netherlands
Indeed the Apple Neural Engine is 16-bit floats under the hood, which gives it extremely limited precision. Plus there is overhead involved in scheduling work to be performed on it (just like on a GPU), and so the CPU is faster for a lot of machine learning models (up to a certain size).
My audio programming blog: https://audiodev.blog
- KVRAF
- 8828 posts since 6 Jan, 2017 from Outer Space
There is this from Apple:
https://machinelearning.apple.com/updat ... ing-on-mac
https://developer.apple.com/metal/tensorflow-plugin/
And more about Audio in Tensorflow:
https://towardsdatascience.com/audio-pr ... 8f1a4103aa
https://www.tensorflow.org/io/tutorials/audio
https://machinelearning.apple.com/updat ... ing-on-mac
https://developer.apple.com/metal/tensorflow-plugin/
And more about Audio in Tensorflow:
https://towardsdatascience.com/audio-pr ... 8f1a4103aa
https://www.tensorflow.org/io/tutorials/audio
Last edited by Tj Shredder on Wed Jun 15, 2022 4:37 am, edited 1 time in total.
- KVRAF
- 8828 posts since 6 Jan, 2017 from Outer Space
Double
-
Music Engineer Music Engineer https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=15959
- KVRAF
- 4292 posts since 8 Mar, 2004 from Berlin, Germany
Having the code available and embedded in a popular open source synthesizer project would actually make it much more appealing to me to try it for some actual sound design because that would give some peace of mind about the future-proof-ness of any interesting results that may come out of such explorations. Also, being able to tweak stuff at the code level or just exploring how it works can be more fun than dealing with a black box.QuadrupleA wrote: ↑Sun Jun 12, 2022 3:07 amGood suggestions for open sourcing or Surge / VCV rack module, would be good do that before I abandon it.