• TalkBass has been independent since 1998. Add your voice.
    Create a free account to reply to discussions, view embedded media, and browse with fewer display ads.
    Join free Log in
    Want zero display ads or expanded classifieds tools? Compare plans.

Double Bass AI and the EUB

Jaygetsreal

Jaygetsreal

Nov 17, 2017
503
1,031
Adelaide South Australia
#1
I think it's fair to say that EUBs and synths have a tough time trying to successfully reproduce the sound of a double bass.
But given the AI processing leading to Now and Then (whether you like the result or not) it's probably now only a matter of time before this technology is embraced by EUB manufacturers. Who'll be first? Yamaha? NS?
Will it be a huge improvement? Whatever, I look forward to the day.
 
  • Like
Reactions: Thumpa
#2
It's still without AI, but with something like the ToneDexter manufacturers could have gone there already (but didn't AFAIK!). Some (thinking MK) seem to have an impressive sound without it anyway, but I agree -- someone will go there and others will follow. Some sooner, some later.
 
#5
Steven Ayres said:
This is an interesting thought, but looking at it in practical business terms, I think aiming this sort of technology at the standard electric bassist makes a lot more sense: a need that's often expressed in a vastly larger market.
Click to expand...

From a business point of view, you are almost certainly correct. However, I would argue that the sound of the Double Bass is far more complex and difficult to replicate.
 
#7
rickwolff said:
From a business point of view, you are almost certainly correct. However, I would argue that the sound of the Double Bass is far more complex and difficult to replicate.
Click to expand...

What AI would bring to it is smarter sampling of those complex waveforms, triggered by and mixing into the tones of the instrument. It actually doesn't sound like a particularly difficult problem relative to, say, deepfaking the Beatles. I agree that EUB would require less processing because you already have the longer strings and what they bring to it, but I don't see that as significantly altering the core challenge, because pickups vary so much.
 
#8
rickwolff said:
I would argue that the sound of the Double Bass is far more complex and difficult to replicate.
Click to expand...

In 1984 I was working the sales floor at LaSalle Music ("May put you on hold please?") when they got the Gen 1 Kurzweill keyboards in. The piano player / sellers were playing the Grand With DB patch and it wasn't quite right.

I played it. Fooled 'em every time.

PBKAC as they say (although they're usually talking about a different kind of keyboard). If somebody who has spent decades learning how to get a DB to sound like a DB spends a little time on it they can create an accurate, idiomatic DB part. It was true in 1984 and it's not less true today.
 
#9
rickwolff said:
Can I get mine with autotune?
Click to expand...

Hmmm, ask Gibson about autotune (aka robot tuners). Their experience involved some heavy equipment...

upload_2023-11-5_9-57-31.jpeg
 
#10
Why stop at EUB/Double bass?
An AI could make any bass (or any instrument for that matter) a more fluid than MIDI type controller which could put any instrument voicing at my fingertips if installed (or outboard) from let's say a Steinberger broom.

When I bought the Zoom A3 it was specifically for the upright model that's within the unit. Does it really make a non upright sound like an upright? To about 50%.

The AI argument takes away everything we ever considered about what it means to actually play an instrument.
What makes anyone think they'll even need us to play, sing, or compose anymore once the machine is truly up to speed?
 
#12
I’m no expert, and I’ve never even played an EUB, but it seems to me that, given enough training data from real upright sounds, AI techniques could be used to build models that would accurately reproduce the subtle variances in sound that come from an acoustic instrument played in a wide variety of techniques and situations, giving a more realistic acoustic sound than is typical of EUBs to date. I don’t know if there’s enough commercial potential to justify the effort it would take.
 
#13
AI is not reliable. You never know what comes out and it is impossible to proof if the machine learned in the same way a human would learn.
Even though most output might be what is expected, something totally unexpected could be generated by the machine.

AI is a dangerous hype because people think that it is a reliable and better solution. It is not.
 
  • Like
Reactions: Tom Lane, Martin Spure and Lance Bunyon
#14
DoubleMIDI said:
AI is a dangerous hype because people think that it is a reliable and better solution. It is not.
Click to expand...

In the present tense I agree with you. But with the mountains of money moving down that lane to meet corporate demand, I expect the sector will be very different in not too long, including with applications we can't anticipate. Appropriately adjusting the timbre of an audio signal seems pretty simple in that context.

I wouldn't drive on a freeway in a Model T, but I also wouldn't underestimate how cutting-edge engineers with real funding can find quick and effective improvements.
 
  • Like
Reactions: Jaygetsreal
#15
I’m heavily into using sampled instruments in Logic Pro. I’ve yet to hear a (real) sampled (or physical modelling) double bass virtual instrument sound like a real one. I’ve tested them all, and some come close, but not close enough. If AI can somehow truly model a double bass in real time, the VST instruments would be the first place they’d appear, as the market is much bigger. And none have appeared yet. There’s not even rumors or gossip about it. That goes for any current virtual instrument, for that matter. They are either samples or physical modelling. And all of them MIDI.

IF such a realtime AI VST instrument is produced, and let’s pretend some clever fella manages to put this in a box you can plug your EUB into, or embedded in the EUB, you’d still need that box to convert the realtime audio from your EUB into MIDI, then processed through the ‘AI VST’ software in the box, THEN converted back to audio. This will inevitably introduce latency.

On top of that, the MIDI ‘tracking’ of the initial realtime audio of the EUB, would have to be second to none, and is notoriously bad in the bass register. Currently Jam Origin – Audio to MIDI is the best we got for audio to MIDI tracking. Also see LoFiLeif’s video below; a highly respected resource in the audio to MIDI world.

Even with AI, so far no realtime ‘turn any audio input signal into another instrument sound’ exist. At least not without having MIDI in the equation (which WILL introduce latency). Believe me, if such software existed, the (huge) Virtual Instrument communities would be ablaze. I don’t think we are close to that scenario for quite some years to come, even with AI. I’ll probably be proven wrong, eventually.. As we all know : It’s hard to predict - especially about the future :)

 
Last edited:
  • Like
Reactions: Jaygetsreal
#16
40 years ago when I began a career as a professional IT systems programmer, AI had already been around in concept for about 30 years, and was discussed often, but never became a real usable tool. There were some programming languages that came out during my career that supposedly made AI programming easy, and they were fun to play with but were all more hype than functionality -- they were geek toys.

There has been progress, but for the common computer-active person today, AI is still just a toy. Maybe we're at a threshold, I cannot claim to know what people are working on. But it isn't here yet in any of the futuristic forms that people have dreamed of for AI in the past.

What I can say from having been around it with geek toys, is that AI at the ground level is basically the same as any other tools in programming. It uses data, it uses if-then processing, and it produces some output -- just like digital computers have done since their inception in the 1940s. The only difference with AI now is that there are a couple of magnitudes more data, an equal increase in processing, and there are hopes and rumors that the output will someday be more useful. But incorporating AI into a search engine or a textual project generator is still basically a toy.

AI gets real is when it is substantially embedded into everyday common devices like cars, phones, home computers, microwave ovens, refrigerators, and into industrial level computers that supply necessities for businesses, homes and individuals. That isn't happening yet -- just about everything noticeable that is AI is online and suffers from the same availability issues as anything else online. So there are still some big fundamental changes that have to occur.

Regarding music, I've heard some pretty impressive digital productions of instrument sounds, but they all come from real instrument sampling at some point. That's going to be very hard to avoid, even for AI. Somewhere in all the data that AI uses for creating music, there will have to be the original sounds created by real instruments; without that it will just be simulation, which is what we already have today.
 
Last edited:
  • Like
Reactions: Anthony White, Martin Spure and Jaygetsreal
#18
Skills and abilities that are not practiced eventually become deficits...
The crucial question not only in this case seems to be, which abilities and skills mankind considers important and worthy of behavior...
By the way....The term "artificial intelligence" is misleading and conjures up the wrong concepts and developments...
This is not only my impression, but also the statement of some scientists who are working on the development of this technology, which certainly also offers many possibilities and opportunities
 
  • Like
Reactions: Martin Spure

Latest posts

Top Bottom
Back