Can ChatGPT make a good preset that emulates a known tone? (Spoiler: no)

tysonlt said:
At least with chatgpt we can skip that charade and just go straight to the source for our inaccurate garbage, and skip the farce of googling.
Click to expand...
ChatGPT often spews inaccurate information because it's not connected to the internet and thus can't check its answers against actual sources; however, that's already changing. Bing with GPT is proving to be far more accurate. In fact, according to one source, it's more than 98% accurate. Further, Bing provides users with the sources it bases its answers on, so users can evaluate the trustworthiness of a source for themselves.
 
Jason Scott said:
Not necessarily. For consumers, art is a means to an end; it arouses emotions and is often a source of enjoyment. In that regard, I don't think AI art defeats the purpose. I mean, I listen to music because it makes me feel good. If AI were able to generate music that made me feel as good as I do when I listen to my favorite artists, I wouldn't let the fact that it was created by AI stifle my enjoyment.
Click to expand...

Fair enough. But, has that happened?

I've spent some time with soundraw.io to see what it could do....technically speaking it was great. But...you could play any of the exact songs I generated for me again, and I'd bet money that I wouldn't be able to tell that I'd heard them before.

I'm really not worried about it.
 
tysonlt said:
When chatGPT writes an essay, it uses sources too - it just makes them up. No joke.
Click to expand...
That's why I wouldn't use ChatGPT to write an essay I needed to be factual. However, if you ask Bing with GPT to write an essay, it cross-checks its output against actual sources.
 
marsonic said:
Fair enough. But, has that happened?
Click to expand...
Not necessarily, but it's definitely close. The best I've heard thus far is from OpenAI's Jukebox; a neural network that can generate music, including singing, as raw audio (in mono) in various genres and artist styles. A few examples:




 
Last edited:
Jason Scott said:
Not necessarily, but it's definitely close. The best I've heard thus far is from OpenAI's Jukebox; a neural network that can generate music, including singing, as raw audio (in mono) in various genres and artist styles. A few examples:




Click to expand...


Uhh....

We are not hearing the same things. Those all sound like absolute garbage. I honestly went and skipped through a playlist of my reference tracks because I thought something was broken.

I went as far as checking the code on the SC page to see if I could find the actual files....and couldn't before I decided it wasn't worth my time. The stuff I'd already written off (as being pointelss) from soundraw.io is worlds better than those examples.
 
marsonic said:
Uhh....

We are not hearing the same things. Those all sound like absolute garbage. I honestly went and skipped through a playlist of my reference tracks because I thought something was broken.

I went as far as checking the code on the SC page to see if I could find the actual files....and couldn't before I decided it wasn't worth my time. The stuff I'd already written off (as being pointelss) from soundraw.io is worlds better than those examples.
Click to expand...
They're two very different architectures. Soundraw's sound quality is undoubtedly superior. However, that's because it's merely generating the composition, not the instrumentation, or at least I haven't read anything that would indicate it generates the instrumentation from scratch and I'd bet the farm it doesn't. To my ears, it sounds a lot like General MIDI. Not bad, but hardly radio-ready.

Jukebox, on the other hand, generates everything from scratch, including the vocals, and it can even incorporate custom lyrics. Granted, it's not radio ready either, but there's plenty of room for Jukebox to improve their model(s), whereas with Soundraw's current architecture, it'll never be anything but the equivalent of a glorified MIDI music generator, in my opinion. Further, Jukebox can generate songs based on specific artists rather than a fairly generic list of styles.
 
Last edited:
Jason Scott said:
I agree, the results are just bad. I asked Bing again today and received this response:

View attachment 117796
Click to expand...

I'm terrible at making presets but just ear-balling it, I started using the Chat GPT instructions and tweaked until it sounded right to me....which again means nothing because I'm the 2nd worst preset creator in the world...

Started with a modern gate in front, default settings.
TS808 full drive, flat tone.
GEQ in front with just a smidge of 4k and 60, just BARELY
FAS Modern 2, instead of FAS Modern 3 worked better.
Gain about 7.5 which is just stupid with the TS808 in front, rest of the tones noon basically, little less bass, tiny more treb NO mid drop.
In the amp block, I actually used a Mark 5 band EQ with nothing boosted, but cut 750 obviously a few points and a very tiny V shape to it, not Master of Puppets kind of V shape.
Cab I used a 57 and 121 mic Marshall (not boogie) and preamp hi/low pass 60 and 6.5k

The weird part....
After the cab...I added another GEQ and started mucking with boosting 4k a little, dropping beyond that highs a little less, some slight cuts in tones that just didn't seem to fit (just again ear-balling it). Probably a PEQ would just nail the right frequencies like 5k.
THEN...ANOTHER modern gate........ yeah, I know, shot in the dark.

Seemed to sound pretty decent but listening to Cowboys from Hell and getting just the guitar part from Moises.ai it sounded close, probably needs a little brightness and some double tracking w/ compression to nail it.

Just my $.02
 
tysonlt said:
There is already so much garbage on the internet. At work I might google a programming question and I reckon in the past year there’s been an exponential increase in pointless fluff articles that clearly don’t understand the topic they’re “teaching.” They’re only there to produce content for its own sake to serve the almighty algorithm, and it’s very annoying. Now people can just get their “blog” automatically generated, and the internet will fill up with ghost-written, inaccurate garbage. Perhaps it is already. At least with chatgpt we can skip that charade and just go straight to the source for our inaccurate garbage, and skip the farce of googling.
Click to expand...
Yes, it's been a long time since I've been able to reliably find answers to programming questions on the internet. Everything brings up GeeksForGeeks.com or some other BS website that gives half-baked and often completely incorrect information.

Honestly it's one reason why I still prefer to stick to forums, discord servers, and other social media. It doesn't take long to suss out who the experts are in a given community (eg: Leon Todd, Yek, etc.) and then you know who you can trust to give reliable information. You can even ask them questions! I think our society is trying to become too reliant on automated systems to answer all our questions and we are losing that connection with with other humans, and respect for experts.
 
Unboxed said:
I'm terrible at making presets but just ear-balling it, I started using the Chat GPT instructions and tweaked until it sounded right to me....which again means nothing because I'm the 2nd worst preset creator in the world...

Started with a modern gate in front, default settings.
TS808 full drive, flat tone.
GEQ in front with just a smidge of 4k and 60, just BARELY
FAS Modern 2, instead of FAS Modern 3 worked better.
Gain about 7.5 which is just stupid with the TS808 in front, rest of the tones noon basically, little less bass, tiny more treb NO mid drop.
In the amp block, I actually used a Mark 5 band EQ with nothing boosted, but cut 750 obviously a few points and a very tiny V shape to it, not Master of Puppets kind of V shape.
Cab I used a 57 and 121 mic Marshall (not boogie) and preamp hi/low pass 60 and 6.5k

The weird part....
After the cab...I added another GEQ and started mucking with boosting 4k a little, dropping beyond that highs a little less, some slight cuts in tones that just didn't seem to fit (just again ear-balling it). Probably a PEQ would just nail the right frequencies like 5k.
THEN...ANOTHER modern gate........ yeah, I know, shot in the dark.

Seemed to sound pretty decent but listening to Cowboys from Hell and getting just the guitar part from Moises.ai it sounded close, probably needs a little brightness and some double tracking w/ compression to nail it.

Just my $.02
Click to expand...
I tried mucking around with the Bing settings in my earlier reply and it was pretty apparent that the results were likely going to sound awful after looking at the actual pre and post PEQ settings. I mean, +12 dB at 100 and 200 Hz for both pre and post EQ? Absolutely horrid.
 
Jason Scott said:
I tried mucking around with the Bing settings in my earlier reply and it was pretty apparent that the results were likely going to sound awful after looking at the actual pre and post PEQ settings. I mean, +12 dB at 100 and 200 Hz for both pre and post EQ? Absolutely horrid.
Click to expand...
LOL...Yeah, honestly I could just mic up my toilet on a saturday morning after beer and spicy mexican food friday night and it would sound pretty close
 
Jason Scott said:
They're two very different architectures. Soundraw's sound quality is undoubtedly superior. However, that's because it's merely generating the composition, not the instrumentation, or at least I haven't read anything that would indicate it generates the instrumentation from scratch and I'd bet the farm it doesn't. To my ears, it sounds a lot like General Midi. Not bad, but hardly radio-ready.

Jukebox, on the other hand, generates everything from scratch, including the vocals, and it can even incorporate custom lyrics. Granted, Jukebox isn't radio-ready either, but there's plenty of room for Jukebox's models to improve and evolve, whereas with Soundraw's current architecture, it will never be anything but the equivalent of a glorified MIDI generator, in my opinion. Further, Jukebox can generate songs based on specific artists rather than a simple generic list of styles.
Click to expand...

These comments literally make zero sense to me.

Soundraw sounds "radio ready" except for level and the fact that they expect you to add elements to it rather than just publishing it as-is. And the music has no emotion.

The clips you posted sound like something involved in creating them was broken. They can't be saved.

We might just have to agree to disagree on this one.
 
marsonic said:
These comments literally make zero sense to me.

Soundraw sounds "radio ready" except for level and the fact that they expect you to add elements to it rather than just publishing it as-is. And the music has no emotion.

The clips you posted sound like something involved in creating them was broken. They can't be saved.

We might just have to agree to disagree on this one.
Click to expand...

Do you seriously think the distorted guitars in their rock category are 'radio ready'? In my opinion, they're about as radio ready as a dial-up internet connection.

I mean, some of their acoustic guitars and piano variations aren't 'bad', but radio ready? In my opinion, no. Like I said, it sounds like glorified General MIDI to my ears. Are you familiar with General MIDI (GM)? Again, Soundraw's sound quality is superior to Jukebox at this stage, but that makes sense given Soundraw isn't generating its instrumentation from scratch, whereas Jukebox generates everything from scratch, which gives it infinitely more flexibility, and once the technology matures, it'll be the equivalent of Midjourney or DALL-E, whereas Soundraw is basically on rails and very limited in terms of what it can and can't do. I mean, there's no option to generate compositions in a given artist's style, nor does it generate vocals, and it's also limited to the instruments shown in their drop-down list.
 
Last edited:
Jason Scott said:
Do you seriously think the distorted guitars in their rock category are 'radio ready'? In my opinion, they're about as radio ready as a dial-up internet connection.
Click to expand...

Compared to the examples that made me dive for my references because I thought something was broken....
 
list all amps available in the axefx 3
Click to expand...

The Axe-Fx III includes a wide variety of amp models, including:
  1. '59 Bassguy
  2. '59 Tweed Lux
  3. '62 AC Boost Vibrato
  4. '62 Brown Bass
  5. '65 Bassguy
  6. '66 AC Hi Boost
  7. '66 AC Hi Boost Vibrato
  8. '66 AC Normal
  9. '66 AC Normal Bright
  10. '66 AC Vibrato
  11. '66 Bassguy
  12. '67 Blackface Bassman
  13. '67 Blackface Deluxe
  14. '67 Blackface Dual Showman
  15. '67 Blackface Pro Reverb
  16. '67 Blackface Princeton Reverb
  17. '67 Blackface Twin Reverb
  18. '68 Plexi Lead
  19. '69 Hiwatt Custom 100 DR103
  20. '69 Hiwatt Custom 50 SA212
  21. '69 Plexi 50W Blues
  22. '69 Plexi 50W Jump
  23. '69 Plexi 50W Normal
  24. '69 Plexi 50W Variac'ed
  25. '70 Super Lead 100
  26. '72 Orange OR120
  27. '75 Dumble Overdrive Special
  28. '78 PV-5150
  29. '80s JCM800 2204
  30. '82 Lead 800
  31. '83 JCM800 2203
  32. '85 Mark IIC+
  33. '89 SL-100
  34. '90 Brit J-800
  35. '90s Solo 100
  36. '92 Treadplate Modern
  37. '96 Matchless DC-30
  38. '99 IIC+
  39. 5150III 50W
  40. Ace Thirty
  41. Angle Fawn Jump
  42. Atomica
  43. B-Killer
  44. BE 100
  45. BE100 Deluxe
  46. BE101
  47. Big Verb
  48. Blackface Dbl
  49. Blackface Dbl Norm
  50. Bogfish
  51. Brit J-45
  52. Brit J-45TB
  53. Brit Plexi Brt
  54. Brit Plexi Jump
  55. Brit Plexi Nrm
  56. Brit Super100
  57. CAA PT100
  58. Cameron CCV100
  59. Cheetah
  60. Citrus
  61. Class-A 30W
  62. Class-A 30W Brt
  63. Class-A 30W Jump
  64. Class-A 30W Nrm
  65. Class-A 30W Vib
  66. Class-A 30W Vol
  67. Class-A 30W XFMR
  68. Class-A 50W
  69. Class-A 50W Brt
  70. Class-A 50W Jump
  71. Class-A 50W Nrm
  72. Class-A 50W Vib
  73. Class-A 50W Vol
  74. Class-A 50W XFMR
  75. Classic Brit 30W
  76. Classic Brit 45
  77. Clean Custom
  78. Custom Audio Amplifiers OD100
  79. Custom Z
  80. D-Zilla
  81. Das Benzin Mega
  82. Diesel VH4
  83. Dual Rectifier Red
  84. Dual Rectifier Treadplate
  85. EL84 Brit 20W Brt
  86. EL84 Brit 20W Nrm
  87. EVH 5150III
  88. Emperor 4x12
  89. Emperor 6x12
  90. Emperor OS
  91. Eng
 
You must log in or register to reply here.
Back
Top Bottom