Forget sampling!
-
I asked ChatGPT to write a presentation of FaustGPT , here is the answer:
FaustGPT is a DSP program generation model for audio built using the Faust language. Faust is a programming language dedicated to creating digital audio processes, such as guitar effects, synthesizers, or real-time audio processing tools. FaustGPT uses machine learning to learn how to generate DSP programs from input-output examples, allowing users to quickly create custom audio effects without having to write the code themselves.
FaustGPT uses a GPT (Generative Pre-trained Transformer) type neural network architecture, which is pre-trained on a large amount of audio data to enable faster and more accurate generation of DSP programs. Users can provide input-output examples for specific audio effects, such as reverb or distortion, and the model will automatically generate the corresponding DSP code.
FaustGPT allows musicians and sound engineers to create custom audio effects without having to learn the Faust programming language or spend time writing DSP code. This makes the creation of new effects more accessible and faster, allowing users to focus on creating unique music and sound experiences.
-
Some progress: http://www.vaporwav.in
-
-
@sletz Aha that's clever
-
@sletz very cool. I'm gonna test this out tonight
-
Do you guys think that this will indeed get rid of us all? Or is there still a way to stand out and maintain sales? I can't tell if these comments are sarcasm or everyone is indeed worried about the sampling business at large.
-
@virtuscapeaudio I don't see it replacing sampling any time soon. Most of the physical modelling algorithms have been around since the 80s and haven't changed much. Wind instruments, plucked strings, and percussion models tend to sound the most realistic but the advantages over sampling are minimal.
I used to think physical modelling was the way to go but it only gets you to the point where you have a sound. To make it useful you still need to do all the scripting stuff you would with sampled sounds, and sample sounds are more realistic. The only real advantages are that it can be more precise than sampling and more dynamic. You don't have to worry about phasing between velocity layers. And generally it uses a lot less RAM and disk space, but is heavier on CPU.
With lazy loading (purge until played) and modern massive hdds/ssds the RAM and disk space for a sample library isn't really an issue these days.
So while I still like the idea of physical modelling, and not having to edit 1000s of samples. In the long run if I want a realistic sounding instrument the fastest way to get there is just to record it.
-
@d-healey Great! I share your sentiments here. ....and I just loaded the HISE snippet from @Christoph-Hart and burst out laughing out loud haha. I needed that and I indeed just got Rick Rolled lololol
-
@virtuscapeaudio said in Forget sampling!:
Do you guys think that this will indeed get rid of us all? Or is there still a way to stand out and maintain sales? I can't tell if these comments are sarcasm or everyone is indeed worried about the sampling business at large.
Yes, old technologies may not have changed much. But the story of artificial intelligence is much more than this. With today's computers, a person's ability to complete a task that would take months or even years in a very short time, using another previously trained software with its own intelligence, is a unique situation in human history.
To be frank, it would be wrong to be too optimistic. In my opinion, within 10 years at most, at least more than half of the work (business) currently done by humans with computers will be destroyed by artificial intelligence. And unfortunately, software development comes first among the tasks related to computers.
We are now witnessing the infancy of artificial intelligence. We will also see this technology stand up and start doing real work instead of people.
There is no escape from this