KugelAudio Launches Real-Time Text-to-Speech Model with Self-Hosting Option

KugelAudio, a new text-to-speech platform developed by a four-person team based in Berlin, launched today on Product Hunt. The service offers real-time TTS capabilities with voice cloning and sub-60 millisecond latency, deployable either on-premises or via API.

The platform distinguishes itself through grammar-aware normalization that intelligently reads phone numbers, IBANs, addresses, and medications naturally across more than 25 languages. Additional features include word-level timestamps, IPA (International Phonetic Alphabet) support, and pre-built integrations with communication and voice infrastructure platforms including LiveKit, Pipecat, and Vapi. Users can also customize pronunciations through dictionary features, allowing for handling of mixed-language contexts such as German text with English product names.

Developers have indicated that European languages are currently the primary focus, with the team actively gathering diverse voices and accents. While the platform lists support for languages including Hindi, stable support for these languages remains under development. The service includes a multi-context endpoint compatible with ElevenLabs SDK integration, particularly utilized in the LiveKit integration.