KugelAudio

Companies building voice bots and call-center automation need natural-sounding speech and accurate transcription, but many leading voice AI providers are US-based, raising GDPR and data sovereignty concerns for EU/EEA customers. Voice bots also commonly mishear names, addresses, IBANs and phone numbers during calls and are forced to guess rather than confirm, creating friction and errors in real customer interactions. The page frames this as an unsolved problem by naming specific competitors (ElevenLabs, OpenAI, Speechmatics, Deepgram) and showing accuracy gaps on names and addresses as low as 49-65% for those providers.

KugelAudio offers a text-to-speech model (Kugel 3 and a cheaper Turbo tier) with claimed 91ms median time-to-first-audio, voice cloning, IPA notation and precise handling of numbers, emails and phone numbers, plus a separate speech-to-text product, Kugel Ear 1, that flags when it is unsure of a word (e.g. a caller's name) and sends alternate spellings so the bot can ask once instead of guessing. A third product, Clarity, removes background noise and other voices from calls and recordings. All models are built, trained and hosted in the EU, with an EU Sovereign Cloud option using only EU-owned infrastructure (explicitly excluding AWS, Google Cloud and Azure) to avoid US CLOUD Act exposure, and an on-premise enterprise deployment option using the same SDK/API. Pricing is per audio minute, billed by the second, with published self-serve rates and a separate custom Enterprise contract for on-premise hosting, SLAs and committed rates.