ai-coustics

ai-coustics builds an audio intelligence layer for Voice AI systems: real-time, on-device speech enhancement that turns raw, unpredictable sound into stable input for downstream voice pipelines. The company was founded in 2021, spun out of Technical University Berlin, and works from Berlin on problems spanning acoustics, signal conditioning and machine learning inference.

Its model family covers the main stages of a voice pipeline. Quail is a speech enhancement model designed for sub-40ms latency and on-device capability. VAD provides voice activity detection, Rook handles multi-speaker audio, and Tyto offers audio insight and analysis. These are delivered through a developer-facing SDK and APIs that integrate into Voice AI stacks; the company reports processing millions of hours of audio each month through them.

The engineering focus is on production properties of voice systems: low-latency processing, on-device inference, and the reliability, observability and measurability of audio pipelines. Customers named by the company include PolyAI, Synthesia, Perplexity, Retell, LiveKit and AssemblyAI, with a global customer base across conversational AI, speech technology and real-time communications.

The team comprises researchers and engineers working across audio, acoustics and machine learning, in a research-driven culture with academic roots at Technical University Berlin.