Roland Future Design Lab has introduced Tone Explorer, a new tool that’s designed to streamline sound patch selection using Artificial Intelligence.
With modern sound libraries growing into the thousands and even tens of thousands of patches, it can be time-consuming to identify the sound that you want to use for a specific part. Tone Explorer uses neural networks to analyze the characteristics of a recorded MIDI phrase and presents an array of matching options, that they say “range from safe to surprising”.
Tone Explorer is being introduced as a functional, but unfinished, technology preview for musicians to try out. Roland says that it’s also an opportunity for early adopters to directly influence future Roland development.
Features:
- Experimental feature designed to assist the creative process
- Uses advanced neural networks to match tones to phrases
- Roland Future Design Lab technology preview for GALAXIAS
- Developed and designed in collaboration with Qosmo
- Requires GALAXIAS Super Instrument and ZENOLOGY Synthesizer
The Tone Explorer Technology Preview is being offered through Roland Cloud as a “GALAXIAS Labs Experiment.” Existing Roland Cloud Ultimate members have immediate access to the preview, while newcomers can sign up for a 3-month trial by visiting the Tone Explorer information page and completing a brief onboarding questionnaire.
“When Roland and Universal Music Group established The Principles for Music Creation with AI, hearing from musicians as we consider and develop AI applications was made fundamental. With the Tone Explorer technology preview, we are honoring this as we seek input from innovative music creators on everything from initial concept to execution,” notes Roland Future Design Lab head Paul McCabe. “If we move forward with Tone Explorer development in the future, it will be based directly on the feedback we receive over the coming weeks.”
More information on the Tone Explorer Technology Preview is available at the Roland site.
What they don’t tell you is that there is a fundamental problem with neural network AI. Essentially, it learns by example. So if the examples it learns from are maudlin, deriviative, tuneless, etc, then it’s going to create similarly bad quality. It’s all well and good when it’s learning from talented, imaginative humans. But the more people use AI, the more it encounters AI examples, and it learns from content it itself created. And like some form of electrical inbreeding, things start to get very unsophisticated, pedestrian, and just plain awful quickly. This is _not_ the future, folks.
It seems like there could be some universally useful rules – like if you are playing a bassline, it would suggest bass sounds, and if you’re playing a lead, it would suggest lead sounds.
But I’d agree, it seems like going much beyond that would lead to ‘lowest common denominator’ patch suggestions.
“maudlin, deriviative, tuneless, etc, then it’s going to create similarly bad quality.” – i.e. Techno and Acid.
btw, there is no future, only now, now, now…. oops, another now.
I don’t see much of a difference between this and using presets or genre-focused virtual instruments and sample packs. If sound design isn’t your thing that’s fine, but then you should probably make sure you’re putting a bit more effort into composition.
nah man… Roland should be doing a new V-Synth with AI powered resynthesis
A new V-Synth powered with a Raspberry Pi (ala Korg) with 32gb SAMPLE RAM + 128gb sample storage onboard. That would be faster and more effective for the V-Synth synthesis which is cool, but which was undermined by slow CPU and memory limits.
Amusing, but part of my fun resides in testing sounds for a part from a hefty library I’ve built over the years. I won’t even remotely automate that. Maybe it can be useful to pros who are working to a deadline, for example. I’ll keep reading as it develops.
As it is, my first reaction is “Oh bugger, another lazy way to let the machine do too much of the heavy lifting.” I’m with J-e-f-f-g. The manner in which AI is “fed” at the moment is via either inbreeding or plagiarism. I’m still waiting to see a more original way to apply it.
“Ah, dubstep. The sound of living in your parents’ basement forever.” 😛
~ “American Dad”
Roland should use that AI of theirs to come up with less cackhanded business practices …. cuz wtf Roland should be bigger than everyone by this point only they can’t find their LP filter cutoff knob with both hands.
tracewidth for the win! 😛 I saw a world map featuring the market reach of various manufacturers. Behringer is the leader by far, with Roland a respectable second. Other companies have their hot regions, but clearly, low price is a prominent factor. Moog is NOT a leader in China. There’s no mystery to it. Kids, newbies and folks on a budget outweigh workstation buyers by miles.
Besides, in my view, the market seriously needs designers who can devise a more approachable GUI. Micro-knobs and stingy screens with 50 pages to manage probably become e-trash quickly.