Text to speech, text to image, text to video. What else haven't we turned a text into? Ah, now how about text to synth? Let's be honest, synth isn't the first thing that comes to mind when the product team is thinking on expanding the functionality of their AI-powered music tech tool.
Not Fadr, though.
Fadr, an AI-driven stem splitter and remix maker, has introduced SynthGPT, "the world's first text-to-synth engine." This AI-powered plugin, now available in beta, promises to revolutionise the sound design process by allowing users to create unique synth sounds through simple text descriptions.
"SynthGPT can make nearly all types of instruments, including live instruments, actual synths, and more. SynthGPT is just a name actually. In my opinion, 'synth' means any instrument that creates sounds electronically (whether analog or digital). There are two main goals for SynthGPT. One is to create a new type of instrument whose sound is controlled by text. Second, we're working on making it react to MIDI input in new, highly expressive ways. So far we've implemented the first goal pretty well, and the second one is a work in progress. You'll be the first to test new versions if you participate in the Beta access program," wrote Gabe Earle, Fadr co-founder, in an email interview to Kill the DJ.
By describing the desired synth sound using text prompts, such as "dark 80s synthwave arp," users can explore and choose from a palette of 100 synth options generated by the AI engine. SynthGPT also allows users to curate a catalogue of their favourite sounds, so the sound making with AI is kind of personalised. You can listen to the demo previews on their website to see how it works.
As the product page reads, "SynthGPT is an ongoing project. At the moment, SynthGPT's AI engine runs on Fadr's servers, and delivers a one shot to the plugin. The first update will radically improve SynthGPT's performance by introducing an AI engine that runs within the plugin. This will enable SynthGPT to intricately adapt its timbre to notes, rhythms, and dynamics in real time. SynthGPT will also see updates such as traditional controls for fine-tuning sounds, text-controlled AI audio effects, sound quality and diversity improvements, [and] audio prompt based synth sound creation."
SynthGPT is compatible with all major Digital Audio Workstations (DAWs) and operating systems. It's currently available in VST3 format for Windows and in both VST3 and Audio Unit format for Mac users.
If you're worrying about data sources, the team ensures "SynthGPT's sounds are like any other synthesizers sounds—royalty free."
"SynthGPT was trained to understand which words, ideas, and phrases humans like to associate with various different sounds. As a result, it was not necessarily trained on popular music alone, which I think contributes to what makes it special. For example, SynthGPT knows what a car engine sounds like, because its training included sound effects. The advantage is that if you ask for a 'bass that sounds like a car engine', it is able to understand your request," Earle says.
The SynthGPT beta is now open to users, exclusively accessible through Fadr Plus—a subscription service priced at $10 per month or $100 per year.