ai-coustics

ai-coustics was founded in 2021 as a spinout from Technical University Berlin and builds an audio intelligence layer for Voice AI. Its real-time speech enhancement solutions turn raw, unpredictable sound into stable, interpretable input directly on-device, with sub-40ms latency. The company processes millions of hours of audio each month and counts PolyAI, Synthesia, Perplexity, Retell, LiveKit and AssemblyAI among its customers.

The product line is a developer-facing SDK and APIs that condition audio signals and emit reliability telemetry for Voice AI companies worldwide. Underneath sit a set of specialised models: Quail, a speech enhancement model for cleaning and conditioning raw audio; a voice activity detection model; Tyto, an audio insight model; and Rook, for multi-speaker processing.

The technical domains are demanding ones - speech enhancement, acoustics, machine learning, real-time audio processing, on-device inference, voice activity detection and multi-speaker processing - applied to the problem of making voice pipelines observable, measurable and enforceable in production. The team comprises researchers and engineers specialising in audio, acoustics and machine learning, working from Berlin with a customer base that spans the globe.