KugelAudio

KugelAudio is a voice AI platform built around a clear European positioning. Its text-to-speech models are trained and hosted in the EU, the company publishes a GDPR-aligned data processing path, and it states that customer audio, text and voice references are never used for training. For organisations with data-residency requirements, there is also an on-premise deployment where the models run inside your own infrastructure behind your firewall, using the same SDKs and API surface as the hosted service.

The technical pitch is latency. KugelAudio publishes benchmark tables comparing its median time to first audio against well-known streaming voice models, reporting 91 ms for its Kugel 3 model over WebSocket streaming with network round-trip included. It also promotes edge-case accuracy — numbers, dates, currencies, IBANs, phone numbers, postal codes and email addresses read correctly — which matters enormously in practical voice-agent scenarios where a mistyped reference number breaks the interaction.

Beyond synthesis, the platform has expanded into speech understanding. Kugel Ear 1 is positioned as a transcription layer that tells your bot when it is unsure of a word, returning candidate spellings for names and addresses so the agent can ask once instead of guessing. Developer integration is a first-class concern: drop-in support for Pipecat and LiveKit, Python, TypeScript and cURL examples, REST and WebSocket streaming, word-level timestamps, dictionaries and speed control. Payment is per audio minute, billed by the second.