KugelAudio is a text-to-speech engine built for production voice applications. It targets developers who need low-latency audio output — under 60ms — for conversational AI, IVR systems, or real-time voice interfaces. Voice cloning lets you replicate a target voice from a sample and deploy it via API or on-premises.
What sets it apart from generic TTS services is grammar-aware normalization: it reads phone numbers, IBANs, postal addresses, and medication names the way a human would, rather than spelling them out awkwardly. This matters in healthcare, finance, and logistics contexts where mispronounced data erodes trust.
Supports 25+ languages with word-level timestamps (useful for subtitle sync or lip animation) and IPA phoneme output for fine-grained pronunciation control. Ships with ready-made adapters for LiveKit, Pipecat, and Vapi, so it drops into existing voice agent stacks without custom integration work.
Built by a four-person team in Berlin. Suitable for teams that need predictable latency, domain-specific speech accuracy, and the option to keep audio generation fully on-premises.