German voices,
made in Germany.
SONA is now in BotTalk.

When you set up a project in BotTalk, you choose a language and then a voice. If your language is German, there is a new name in that second list.

We want to tell you who built it, what you can set, and what we cannot tell you yet.

Why German text-to-speech is our main job

Most of what BotTalk reads aloud is German. Between September 2025 and August 2026 about 75% of the audio we produced was German.^[5]

The big voice models are built for many languages, and on German news text they still slip in ways your listeners hear immediately. An English sound creeps into a German name. A year comes out as a different year. Once in a while a voice starts to sing in the middle of a sentence.

Our control layer exists to catch those mistakes before your listeners hear them, and it works above every engine we connect. It holds 21,900 pronunciation rules across 15,900 dictionaries, it rewrites numbers and abbreviations the way they are spoken before the text reaches an engine, and it checks for the moments when a model makes something up.

That layer has less to do when the engine underneath already speaks good German. To our ears SONA does, and that is why it is in BotTalk. It sits next to ElevenLabs, Gemini, OpenAI, Azure and Amazon Polly under the same layer. You choose the voice for each project, and you can swap it whenever you like.

Who built SONA

SONA is made by SONALAB GmbH in Potsdam.^[2] The company is part of the MediaTech Hub Potsdam, which sits in Babelsberg right next to the film studios.^[1] Its two founders come at the same problem from two sides. Alexander Wolf comes from sound and sound design, with Bavaria Film and ZDF among his stations. Sören Hübner, the CTO, brings the software and the machine learning.^[1][6]

The company grew out of a research partnership with the Fraunhofer Institute for Integrated Circuits IIS, and it calls itself a Fraunhofer spin-off.^[1] If the institute’s name does not ring a bell, its best-known work will. The mp3 format comes from there. In 1987 the institute formed a research alliance with the University of Erlangen-Nuremberg for it, and in 1995 the researchers voted on the file extension themselves.^[3] When we say German engineering, this is the tradition we mean.

SONA was built for dubbing and post-production studios, and its plugin works inside Pro Tools, Nuendo and Premiere.^[2] In a studio a generated line has to sit in a film mix next to lines that actors recorded, and the people who judge it listen for a living. That is a hard audience to convince.

We also like how the founders talk about their work. In an interview in July they said: “We do not promise that machines sound better than people. We would rather show concretely what works and leave the verdict to the studios.”^[1] The interview is in German and the translation is ours. It is how we want to work with newsrooms, too.

What you can set

In the settings of a project, under Audio, you choose German as the language and then the SONA voice in the list beside it. In our screenshot that voice is Elin, listed as de-DE-Elin-SONA. Under the list you see a star and the line “Premium voices are the highest quality on BotTalk”. The star marks our premium voices, and SONA is one of them. Preview plays the voice.

Two controls under the voice belong to SONA. Speech speed runs from 0.8 to 1.2, and the pitch is kept, so a faster read does not turn into a higher voice. Stability runs from 0 to 1. At 0 the delivery varies the most, and at 1 it is the steadiest.

One more switch sits on the same card, and we want you to know it is there. It is called Fallback TTS provider. BotTalk uses it when a provider refuses a text, and when three retries in a row fail, so that an outage at one provider no longer costs you the article.

A voice from Germany, for readers in Germany

Elin is listed as de-DE, which is German as it is spoken in Germany. That matters more than it sounds, because listeners hear where a voice comes from. When heute.at in Vienna looked for its voice, an earlier evaluation of AWS Polly failed on the accent alone. The CTO, Alexander Ottitzky, gave us the sentence we still quote.

We take that sentence seriously in both directions. For readers in Germany, a voice built in Germany is the right place to start. For readers in Vienna or Zurich, Elin would be the voice from Paderborn, and we would choose a different one with you. You set the voice for each project, so one publisher can run both.

Made in Germany, in practice

Made in Germany is easy to print on a box, so here is what stands behind it in this case. SONALAB is a German company with its seat in Potsdam. It says its platform is built, hosted and processed on EU infrastructure, that its servers run in Germany, and that uploads, processing and renders never leave that infrastructure.^[2] BotTalk runs on infrastructure in Germany as well.

Then there is a question we expect every newsroom to be asked sooner or later: can you show where this audio came from? SONALAB’s answer on its own platform is a tool called SONA Verify. You drop a file into it and see whether it came from SONA, which voices were used and who owns it. The company says it is building the tool to survive cuts and re-encoding into formats such as mp3. When a voice artist joins, the consent is recorded in the same step as the voice.^[2] In the founders’ words, consent and traceability are built in from the start and not bolted on afterwards.^[1]

The law is moving in the same direction. Article 50 of the EU AI Act asks providers of systems that generate synthetic audio to mark their output in a machine-readable way.^[4]

What we cannot tell you yet

We want to be straight with you about the limits. For the engines we have run for years we can show you an error rate with its denominator. For SONA we cannot yet, because it is new in BotTalk. We also will not tell you it is the best German voice on the market, because we have no measurement that proves it. We can show you who built it, where it runs and what you can set. The rest you can judge with your own ears, which is what SONA’s founders ask of their studios, too.

Hear it on your own articles

The quickest way to judge a voice is to hear it read something you published. Write to Andrey at andrey@bottalk.io and send three articles with the names and numbers your current voice gets wrong. We will play them back to you in SONA’s voice and in the voice you use today, and you decide by ear.