tanh lab — audio software agency

We develop audio software and real-time machine learning for audio and video signals. We work with audio software companies, from a first feasibility study through to release.

Audio software development
Plug-ins, standalone applications, mobile and web. VST3, AU and CLAP, in C++ and JUCE. Cross-platform builds, signing and notarisation.
Real-time inference
Neural inference within a real-time audio budget, on device. We develop and maintain anira.
Research and prototyping
Feasibility studies for methods that have not been built before, and prototypes that establish what they cost.
Interface and instrument design
Interfaces developed alongside the engine, for instruments and for tools.
Consulting and technical strategy
Method and architecture review, technology selection, and technical due diligence.

We develop audio software for studios and manufacturers.

Partners

We work at the intersection of audio software, machine learning and instrument design. We conduct applied research for clients, and translate findings into shipped products.

Background

Our roots lie in research, and we are always looking for new problems to solve.

  1. SeamLess: Distributed Spatial Audio Rendering on the Linux Audio Stack

    A modular real-time renderer that spreads spatial audio processing across several Linux servers, so Ambisonics and Wave Field Synthesis can drive hundreds of loudspeakers — past what a single machine can carry.

    Linux Audio Conference 2026

    Fares Schulz, Max Weidauer, Stefan Weinzierl, Henrik von Coler

    Read on HALDocumentation

  2. Pitch-Conditioned Instrument Sound Synthesis From an Interactive Timbre Latent Space

    A two-stage model that disentangles pitch from timbre, so instrument sounds can be steered through a small, navigable latent space rather than the high-dimensional one such systems usually leave you with.

    DAFx 2025 — Int. Conf. on Digital Audio Effects

    Christian Limberg, Fares Schulz, Zhe Zhang, Stefan Weinzierl

    Read on arXivListenGitHub

  3. anira: An Architecture for Neural Network Inference in Real-Time Audio Applications

    A cross-platform C++ library that moves neural inference off the audio callback and onto a static thread pool, with ONNX Runtime, LibTorch and TensorFlow Lite behind one interface.

    IS² 2024 — IEEE Int. Symposium on the Internet of Sounds

    Valentin Ackva, Fares Schulz

    Read on arXivGitHub

We believe the infrastructure under audio software should be shared. Code that others can read improves, and it is easier to trust in something you ship.

Companies using our open source libraries

  1. aniraReal-time neural inference for audio

    High-performance C++ library for real-time-safe neural network inference inside audio plugins. Multiple backends, deterministic latency. GitHub

  2. tanh-libModular C++ audio library

    Four independently buildable C++20 components: threading, lock-free state, DSP, and audio I/O — the foundation under our plugins. GitHub

  3. ScycloneNeural timbre transfer plugin

    Real-time audio plugin that morphs incoming signals into learned target timbres using a RAVE-based variational autoencoder. GitHub

tanh lab is an audio software studio in Berlin, started by Fares Schulz and Valentin Ackva.

Team
Valentin AckvaLinkedInGitHubORCIDFares SchulzLinkedInGitHubORCIDRodrigo DiazLinkedInGitHubORCIDJakob StolbergLinkedInGitHubLina CampanellaLinkedInGitHub
Location
Berlin, Germany

Get in touch

We love a challenge.

We are always on the lookout for new challenges. Please reach out, we would be happy to help.

contact@tanh-lab.com

GitHub ·Legal·Privacy