Many musicians are understandably wary of generative music models. These tools can represent the worst of AI – performing functions that actively infringe upon human creativity, often trained on copyrighted material and doing so in a ‘black box’ that gives users little oversight or control of the process.
Engram, the debut project of developer Thoughtful Things, is something very different though. Described as a ‘generative audio sampler’, this hardware instrument makes use of generative audio models, but approaches them in a way that its creator likens to circuit bending, allowing users to push the models “beyond their limits and discover expressive, uncanny new sounds to sample.”
The basic format of Engram is similar to what you’d expect from a traditional sampler: It features an onboard mic and stereo line-in for sampling, the ability to step sequence and recall patterns, and both MIDI and CV connectivity. Where it becomes more adventurous though is in its incorporation of what its developer refers to as ‘tiny AI’ – a distinction aimed at differentiating its self-contained models from the ‘big AI’ of Suno or ChatGPT.
The incorporation of AI allows Engram to process and warp recordings in interesting ways, aimed at creating uncanny and glitchy textures out of existing audio using a range of tweakable parameters. It can also generate samples from scratch, which can then be treated with the same processes.
Its onboard mic is said to offer both lo-fi sampling and voice control. There’s also an intention to incorporate plugin-style audio models that can be swapped in and out in a manner described by its creator as being like old-school ROMpler cartridges.
“As a researcher and a musician, I’ve been exploring AI/ML and art for a while,” writes Thoughtful Things founder Evan King on Engram’s Kickstarter page. “I’ve trained models to generate uncanny plants and insects for album art; I've coaxed haunting sounds out of music models by glitching their inputs; I've developed poetic songwriting tools that induce hallucinations from speech recognition models.
“Over time, I've developed a hot take: black box AI art is definitionally boring.
“This is because black boxes are designed for efficiency rather than human agency. Efficiency doesn't leave room for experimentation and human imperfection – the things that make art interesting. Today's tools are great for churning out corporate assets, but they're bad for artists: people who are always in search of new ways to expand their distinct creative processes.”
The post goes on to explain that Engram is designed to explore the aspects of AI that King finds interesting, namely pushing the concept beyond its intended uses and into unexplored territory. Thoughtful Things also confirms that Engram’s models are trained entirely on open datasets that only contain audio licensed for commercial use.
Right now, Engram is live on Kickstarter and, at the time of writing, is more than three quarters of the way towards hitting its target of $10,600 with 27 days to go. According to the developer, the instrument itself is currently a playable and manufacturable prototype that is being actively tested by musicians, after which the aim is to create a final, retail-ready design.
Take a listen to demos in the video above. Those interested in backing the project can secure a first run production unit for $770. Head to Kickstarter or the official site for more information.
