KugelAudio

KugelAudio is a Berlin text-to-speech provider that trained its own voice models rather than reselling someone else's. Two tiers are available: kugel-2-turbo at €0.0375 per generated audio minute and kugel-3 at €0.075, both billed by audio duration.

It targets voice agents, so the design priority is latency: the company reports 40-80ms server-side time-to-first-audio, with drop-in support for Pipecat and LiveKit pipelines. Voice cloning, word-level timestamps, speed control and pronunciation dictionaries are included on both tiers.

The team also released kugelaudio-0-open under MIT, so the underlying approach is inspectable. Training and hosting happen entirely in the EU, which is the pitch for teams avoiding US CLOUD Act exposure.