Vienna-based start-up develops new AI tool for music production

A promotional graphic split into two halves. On the left, colourful, flowing 3D soundwave lines in shades of purple and blue appear on a black background. On the right, a white background features the logo of EuroCC Austria at the top and bold black text that reads: “New AI tool for music production trained on the ASC".

An AI tool for electronic music developed with support from EuroCC Austria generates complete tracks and editable project files using fully licensed training data – a true innovation especially for professional producers.

Major US-based players in AI-generated music now face competition from Austria: the Vienna-based start-up Beat Shaper has trained an AI model that supports musicians in songwriting and sound design by generating beats, melodies, basslines, as well as synthesiser and effects settings. Like ChatGPT, the tool creates its output based on text input.

“Beat Shaper supports both absolute beginners, who often struggle with complex music production software, and professional or semi-professional musicians,” explains Taylor Peer, one of the start-up’s four founders. “It simplifies music production at all levels – removing the steep learning curve for newcomers and offering powerful generative features for experienced producers.”

 

Beat Shaper builds an AI model from scratch

With just five million parameters, Beat Shaper is a relatively small model – a small language model, to be precise. But even training smaller models requires serious computing power, especially when working from scratch and testing many different versions.

Unlike most systems trained on finished songs, the team chose a symbolic music approach: the model receives instructions that tell it how a piece of music should be played. This is comparable to a human musician reading from sheet music – not only which notes to play, but also the tempo and rhythm of the piece.

This way, the model learns that it's not only about the notes – many other factors influence the sound. “It’s similar to playing a synthesiser,” explains Taylor. “There are dozens or even hundreds of parameters that shape the sound – oscillator settings, filter configurations, and more. All of that affects how a tone is produced. No one had ever trained a generative model like this before, so no dataset existed. That’s why we had to build everything ourselves."

One of the biggest challenges was avoiding overfitting: Beat Shaper is designed to generate original, creative pieces that go beyond the training data. Its job is to create something new – but not so experimental that it no longer sounds like music. For music to be enjoyable and recognisable, it still has to follow certain rules. The key during training was to balance innovation with structure, without simply recycling existing content.

 

Supercomputer support: hundreds of models trained simultaneously

Achieving this required massive experimentation and computing resources. Working with just a few models wasn’t enough – the team needed hundreds. That’s where Austrian Scientific Computing (ASC - formerly known as Vienna Scientific Cluster – VSC) came in. Beat Shaper trained over 500 individual models on the Austrian supercomputer, regularly selecting and improving the best-performing version.

Thanks to the ASC, training times were reduced by a factor of 50 compared to the team’s own infrastructure. Each model took only one to two hours to train. “The biggest advantage of using the cluster was being able to train so many models simultaneously. Without that, we never could have experimented at this scale, and that would have negatively impacted the final model’s quality,” says Taylor. The best model was continuously fine-tuned on the ASC until its output met the team’s high standards.

Unique: editable music project files and legal certainty
What makes Beat Shaper unique is that it doesn’t just generate finished audio files – it can also produce editable music project files, compatible with standard production software. Its legal foundation is also different from major competitors: all training data comes either from fair use sources or was fully licensed by Beat Shaper for model training. The model was trained exclusively on data explicitly cleared for use. Creators also remain on safe legal ground – they retain full copyright and usage rights to the material they generate and modify using the tool.

 

Taylor Peer on working with EuroCC Austria

EuroCC Austria supported Beat Shaper with free access to the VSC-5 supercomputer. “I think it’s fantastic that the EuroCC programme exists. We’re competing against significantly larger and better-funded companies, most of them based in the US. Given the lack of substantial early-stage venture capital in Europe, programmes like this are crucial for helping European companies stay globally competitive.”

 

About Beat Shaper

Beat Shaper was founded in Vienna in 2024 by Taylor Peer, Jeremy Brown, Joaquín Molina, and Filip Miškovic. The AI co-pilot for music producers integrates generative AI directly into the production workflow, supporting both beginners and professionals in creating electronic music. The idea for the product first emerged in 2023. Thanks to investment from Vienna High-Tech Incubator INiTS, funding from the Austrian Research Promotion Agency (FFG), and technical support from EuroCC Austria, the team succeeded in launching this innovative AI tool – despite intense competition from large US-based companies.

https://www.beatshaper.ai
https://www.linkedin.com/company/beat-shaper