KugelAudio
KugelAudio is a German voice AI platform for developers building voice agents. Its Kugel 3.5 text-to-speech model speaks 39 languages with native accents and reads numbers, IBANs and phone numbers digit-perfect, while Kugel Ear 1 transcribes callers and flags words it may have misheard. Pay per audio minute, billed by the second, hosted in the EU and fully GDPR compliant.
What is KugelAudio?
KugelAudio is a voice AI platform built in Berlin for teams that need speech in and speech out without sending audio to a US cloud. It covers both directions of the problem: text-to-speech models that give a voice agent its voice, and speech-to-text that writes down what callers actually said. Everything runs on European servers, the company publishes an EU data processing agreement, and customer audio is never used for training.
The text-to-speech side has three models. Kugel 3.5 is the default: it keeps every voice in its own accent, handles one-word replies cleanly and reads numbers and codes in the groups they were written in. Kugel 3 is the previous generation, listed at a higher per-minute rate, and Kugel 2 Turbo is the cheap high-volume option that drops IPA notation and some precision on numbers. All three cover 39 languages with native pronunciation, accents and dialects, and the headline latency figure is a median 91 ms to first audio on Kugel 3 over WebSocket streaming, measured with network round-trip included. Voice cloning takes a clean reference recording and returns a usable custom voice in under a minute.
The edge-case work is what KugelAudio markets hardest. Long IBANs, phone numbers, postal codes, policy IDs and email addresses are read digit by digit or in the groups you typed, and the text layer accepts IPA notation, pronunciation dictionaries, break tags and speed control, with word-level timestamps coming back with the audio.
Kugel Ear 1 is the newer product, and it is speech-to-text for voice agents rather than for transcription apps. When it is unsure of a word, typically a caller's name, it returns the spellings it heard with probabilities so the bot can ask once instead of guessing. It also does its own end-of-turn detection, which stops a bot cutting in when a caller pauses in the middle of a customer number. KugelAudio's own benchmarks report 93.4% of words right on support calls, 69% exact on names and addresses, and a 72 ms median to final transcript across nine European languages.
For developers there is a REST and WebSocket API, SDKs for Python, TypeScript, Java, Rust, C++ and Dart, official plugins for LiveKit, Pipecat, Vapi and Cognigy, and an ElevenLabs-compatible proxy for code that already exists. A separate Clarity model strips background noise and other voices before transcription. Billing is per audio minute, charged by the second, with a one-off five-euro credit on signup. Enterprise adds on-premise deployment, SLAs, custom concurrency and contract terms for professional secrecy holders.
KugelAudio Core Features
Natural text-to-speech in 39 languages with native pronunciation, regional accents and dialects, served by the kugel-3.5, kugel-3 and kugel-2-turbo models.
Kugel Ear 1 speech-to-text that returns the words it is unsure of together with alternative spellings and probabilities, so a bot can ask once instead of inventing a name.
Built-in end-of-turn detection (kugelturn) that waits through a caller's mid-number pause rather than treating silence as the end of the turn.
Voice cloning from a short clean reference recording, finished in under a minute and included in the per-minute rate.
Number and code accuracy
IBANs, phone numbers, postal codes, policy IDs and email addresses read in the groups you wrote, backed by IPA notation, dictionaries and an integrated normalizer.
REST and WebSocket streaming with a median 91 ms to first audio on Kugel 3, plus word-level timestamps, break tags and speed control.
Developer surface covering Python, TypeScript, Java, Rust, C++ and Dart SDKs, official LiveKit, Pipecat, Vapi and Cognigy plugins, and an ElevenLabs-compatible proxy.
European hosting by default, an EU Sovereign Cloud tier that excludes AWS, Google Cloud and Azure, on-premise deployment, and a Clarity model that removes background noise and other voices from call audio.
Who is KugelAudio for?
KugelAudio is aimed at the team that owns a voice agent and has a compliance question attached to it: usually a product engineer or platform lead at a European company who has been told that call audio cannot leave the EU, and who now needs a text-to-speech vendor that will not turn the architecture review into a six-month legal project. It assumes you are building or integrating software, not clicking a studio tool, and that you can read a rate card and estimate your own audio minutes. Roles it typically serves: - Developers building inbound phone agents for clinics, insurers, utilities and public services, where caller names, addresses and customer numbers have to come out right. - SaaS and contact-centre teams replacing an ElevenLabs or Deepgram dependency to satisfy GDPR, the EU AI Act or a customer's data-residency clause. - Teams already running LiveKit, Pipecat, Vapi or Cognigy pipelines who want to swap the speech layer without rewriting the orchestration. - Product teams localising a voice experience into German, French, Polish, Czech or Turkish, where accent authenticity matters more than voice variety. - Regulated organisations, including professional secrecy holders such as law firms and tax advisers, that need on-premise hosting and a custom DPA. - Media and e-learning producers narrating long-form content at a per-minute rate with no subscription. - Voice-AI startups that need one vendor for both speaking and listening, so the turn-taking logic can use the uncertainty signal rather than working around it. It is a weaker fit for hobbyists who want a polished browser studio with monthly credits and a preset voice marketplace, for anyone who needs hundreds of celebrity-style stock voices, and for teams whose only requirement is raw transcription of meeting recordings across dozens of non-European languages. Feature breadth is narrower than ElevenLabs, and the interesting parts of the product, word alternatives and turn detection, only pay off if your stack can consume them.
KugelAudio Use Cases
Answer inbound phone lines with a voice agent that asks callers to confirm a name instead of silently mishearing it.
Localise a product, IVR or support line into 39 languages with voices that keep their regional accent.
Read IBANs, policy numbers, phone numbers and postal addresses aloud without digit errors.
Clone a founder or brand voice for consistent narration across ads, onboarding and help content.
Drop European speech into an existing LiveKit, Pipecat, Vapi or Cognigy pipeline without rebuilding orchestration.
Clean up noisy call recordings with Clarity before transcribing or archiving them.
Run an on-premise deployment for a regulated or secrecy-bound organisation that cannot use a hosted API.
Narrate e-learning, audiobook or podcast content at a per-minute rate with no subscription commitment.
KugelAudio Pros and Cons
Pros
- European data residency is the default rather than an add-on: EU hosting, an EU DPA, a subprocessor register and a Sovereign Cloud tier that excludes US-owned hyperscalers.
- Accuracy on the words that break voice agents is measurably strong, with digit-perfect IBANs, phone numbers and codes, and a 69% exact-match score on names and addresses.
- Kugel Ear 1's uncertainty signal is a real design difference: returning ranked spellings lets an agent resolve a name in one question instead of guessing.
- Latency and pricing are both competitive, at a median 91 ms to first audio and €0.035 to €0.0525 per minute of generated speech, billed by the second.
- Integration path is short, with official LiveKit, Pipecat, Vapi and Cognigy plugins, six SDKs and an ElevenLabs-compatible proxy.
Cons
- There are no subscription tiers or seat plans, only per-minute usage plus a sales-negotiated Enterprise contract, so budgeting depends on forecasting audio volume and volume discounts require talking to sales.
- The voice catalogue is a curated set of recorded voices rather than a large marketplace, and the integration list is shorter than ElevenLabs' ecosystem.
- The accuracy and latency claims come from KugelAudio's own KugelBench runs rather than independent third-party evaluations, and speech-to-text covers nine European languages against 39 for text-to-speech.
FAQ About KugelAudio
KugelAudio Pricing
KugelAudio bills purely by audio duration, with no subscription tiers. Every new account gets a one-off €5 Text to Speech credit, at least 70 minutes of audio, with no card and no expiry. After that, text to speech costs €0.0525 per minute on kugel-3.5, €0.07 on kugel-3 and €0.035 on kugel-2-turbo; Kugel Ear 1 speech-to-text is €0.008 per minute (about €0.48 an hour) and Clarity noise removal is €0.0015 per minute of input audio, all excluding VAT. Charges are per second of audio, with a one-second minimum per Clarity request and no charge for failed requests. EU hosting, Sovereign Cloud, voice cloning, streaming and timestamps are included in every rate. Enterprise adds on-premise deployment, SLAs, a custom DPA and committed-volume discounts.
Check official pricingFree credit
€5 of Text to Speech credit on every new account, at least 70 minutes of audio with any model. No credit card required, granted once, no expiry. Speech to text and Clarity are not covered by the credit.
Clarity
clarity-1 removes background noise and other voices from calls and recordings. Billed per second of input audio, rounded up with a one-second minimum per request; failed requests are not billed.
Speech to Text
kugel-ear-1 writes down what callers say, flags words it is unsure of with alternative spellings, and includes end-of-turn detection. About €0.48 per hour, excluding VAT.
Text to Speech Turbo
kugel-2-turbo for high-volume jobs. Lower cost, but no IPA notation and less precision on numbers, emails and phone numbers. EU hosting, cloning, streaming and timestamps included.
Text to Speech
kugel-3.5, the default model, sharper on short replies, pauses and spelled-out codes, with IPA notation, dictionaries, normalizer and precise numbers and emails. €0.062475 incl. 19% VAT.
Text to Speech kugel-3
The previous-generation model, listed at a higher per-minute rate than the newer default. Same feature set as kugel-3.5 including IPA notation and digit-accurate numbers; billed per second of generated audio.
Enterprise
On-premise hosting, SLAs, custom DPA, terms for professional secrecy holders (§ 203 StGB), dedicated account manager, 24/7 support, higher concurrency limits and lower rates with committed usage.
KugelAudio Alternatives
Play HT
Free browser-based text to speech preview that lets you hear your script with system voices before sharing.
Fliki
Fliki is a browser-based AI video and audio creation platform that combines text-to-video editing with synthetic voiceovers, stock media, and generated visuals. Users can repurpose articles or presentations, create avatar videos, translate content, and produce social clips. A free watermarked plan supports evaluation, while paid subscriptions unlock commercial rights, higher resolution, additional voices, and larger credit allowances.
Voice.ai
Voice.ai is an AI voice platform that combines a real-time voice changer, natural text to speech, instant voice cloning, and human-sounding voice agents in one toolkit. It offers a large library of voices and accents, browser and desktop apps, plus developer APIs and SDKs. Free credits let anyone start generating audio or automating calls before upgrading.