[Free] Introduction of TextToSample that can create loops from text with Samplab's AI technology

Introduction of TextToSample, which can create loops from text with Samplab's AI technology

Samplab has released a plug-in "TextToSample" that converts text input into sound.

The experimental elements are still strong, but surprisingly usable sounds are generated.

And the license of MusicGen Model Card isCC BY-NC 4.0So I think commercial use is NG.
(I don't know if the generated sample is also NG)

Sound demo

The official demo video takes less time to generate, but it depends a lot on your CPU specs.

With my PC Ryzen 5 5600G, it takes more than 6 minute to generate a loop of about 1 seconds.

So here are some generated loops, listen to them.

Keywords are "drumnbass" and "EDM".

You can also prepare a different sample at the beginning and generate the continuation.
(Don't expect too much)

In the video, the last loop is generated from the content that was played first, and then set the number of seconds to 8 seconds.

The keyword "dubstep" was a pretty good result.

Warnings

In my environment "Samplab" is also installed, but for some reason "Samplab” started.

As a first aid measure,Samplab"" folder to another location, "TextToSample" started successfully.

If you end up in the same situation,Samplab" folder to another location and try again.

Overview

Since there is only information on the official Q&A rank, I will post the Japanese translation.

Q.Do I need money to use TextToSample?
A.No.

What is the purpose of Q.TextToSample?
A.TextToSample's purpose is to enable the community of producers to experiment with state-of-the-art generative AI models.We want to show what is currently possible with these models, especially when running locally on your computer. TextToSample's goal is not to replace human musicians.

Q.What is the license for TextToSample?
A.TextToSample uses Meta's MusicGen model weights and is licensed under CC BY-NC 4.0.Specifically, the weights of the “small” model are used.Learn more about licensing and intended use here.

Q. What kind of data did you learn from?
A.TextToSample uses weights from Meta's MusicGen model.Learn more about models and the data they are trained on here.

Q.Is internet connection required?
A. All calculations are done locally on your computer.An internet connection is only required to download the model the first time you open TextToSample.

Q. What kind of hardware do I need?
A.We highly recommend Apple M1 chip or GPU with Neural Engine.Otherwise, the model will take a long time to generate speech.

Q.Which OS do you support?
A. Windows: 10 or more
MacOS: 12 and above

Q.Is there a VST2 version?
A. VST2 plugins have been discontinued and can only be distributed by developers who have created VST2 plugins in the past, so unfortunately it is not possible to obtain a VST2 license.If you really need a VST2 plugin, I recommend using a VST3 to VST2 bridge like Kushview's Element.We are not affiliated with Kushview and cannot be held responsible for Kushview's products working with TextToSample.

 

Download

No registration required to download.

Go to the link above and click "Download TextToSample" to download the installer for your OS.

Installation

Win is VST3, so the default installation is fine.

Summary

Compared to graphics, I feel that sound generation AI technology is still far behind.

However, technological innovation is steadily occurring in terms of reproducing effects.

IK'sTONEX,Mr. Tone EmpireA plug-in created by AI technology, and it's freeNeuralAmpModelerPluginSomething is the result of AI technology.

I think there will be a breakthrough in the sound generation system in the not too distant future.