AI in Practice: Four Examples from the Bavarian Audio Industry

1. Antenne Bayern: A chatbot that knows what’s currently on the air

What happens when a radio station gets a chatbot that actually knows what’s currently on the air? This question was the starting point for a feasibility study conducted by the AI Competence Center for Media (KI.M) in collaboration with Antenne Bayern.

The problem with traditional radio chatbots: They know the broadcast schedules, the hosts’ profiles, and the FAQs. But what’s happening live right now is a blind spot for them. The KI.M project solved this with a Retrieval-Augmented Generation (RAG) architecture that integrates live transcripts of the current program with minimal delay. Listeners who want to know which song was just playing or what topic the host just addressed receive a truly up-to-date answer.

At the same time, the team has developed a multi-tiered security system: An upstream filtering model reviews every request from an editorial perspective before it is answered. Politically polarizing questions, requests outside the broadcaster’s context, or attempts at manipulation are detected and blocked. The result is a proof of concept that demonstrates: real-time AI-powered listener communication is technically feasible – and can ensure brand safety.

The evaluation showed that locally operated open-source models can compete with commercial API solutions in terms of quality – an important finding for broadcasters that need to prioritize data sovereignty. The full final report is publicly available.

Karsten Wellert, a member of the executive board and CDO of the Antenne Bayern Group, describes the added value of the collaboration as an outside perspective. In particular, he said the approach of incorporating live transcripts directly as a source of knowledge was new to him.

2. Radio Gong 96.3: From Briefing to Commercial in Minutes

For a long time, radio advertising was a manual process: briefing, copywriting, studio session, voice-over recording, and approval. Radio Gong 96.3 in Munich shows that there’s another way. With the AI-powered ad generator Radio AD Maker, local clients have access to a tool that quickly creates production-ready ad drafts—directly from a simple product description.

For a local station with many small advertisers, this represents a significant gain in efficiency: commercials that used to require time-consuming rounds of revisions are now produced much more quickly. People are still involved in the process – but AI handles the initial draft. This lowers the barrier to entry for smaller advertisers and opens up formats that simply weren’t economically viable before.

3. Fraunhofer Institute: The Voice That Never Sleeps

When an alert is issued, every second counts. And every syllable must be right. It is precisely in moments like these that Allinga TTS from the Fraunhofer Institute for Integrated Circuits (IIS) in Erlangen is already in use. WDR uses the technology for automated text messages in the event of a disaster – especially for listeners with visual impairments who have activated audio description and thus receive warning messages via a synthetic voice.

What sets Allinga TTS apart from many competitors is that the system runs on-premises – on the broadcaster’s own infrastructure, without a cloud connection. A single laptop is all that’s needed to run an instance. That may sound modest, but it has major implications: If the Internet goes down, the system continues to function, and sensitive audio data never leaves the premises. That’s exactly what makes it appealing to public broadcasters, for whom data protection isn’t an option – it’s a requirement.

Allinga TTS was already in use at Bayerischer Rundfunk during Warntag 2025. Additional projects with media companies from Bavaria and the German-speaking world are underway – but cannot yet be publicly disclosed. Valentina Ciardini, Senior Product Manager at Fraunhofer IIS, describes the interest from the industry as growing: “Media companies are gaining a better and better understanding of where the real benefits lie. And as a result, they’re using the system much more intelligently.”

Another feature that is increasingly in demand in broadcasting: Custom Voice. Broadcasters can have their own branded voice trained – based on recordings ranging from 20 minutes to two hours of the actual speaker. The result is not a generic voice, but one that truly resembles the original person. Austrian German is already part of the commercial portfolio – Franconian and other dialects are in development.

4. Goaudio: A daily podcast episode featuring hyperlocal AI news

Local news, fresh every day, produced fully automatically: With the podcast “Servus Niederbayern!,” Mediengruppe Bayern, in collaboration with the AI audio platform goaudio, has established a workflow that demonstrates what’s possible in news production today. The editorial team selects the relevant stories, and goaudio handles everything else: script creation, speech synthesis, sound design, and distribution. The result is released daily as a finished podcast on Spotify and other platforms.

The principle behind goaudio is clear: input news stories, output newscasts. The platform is specifically designed to handle the nuances of the German language, integrates with existing CMS systems via API, and gives the editorial team full control over the final product. According to the provider, this model saves up to eighty percent of production time compared to traditional workflows.

In addition to Mediengruppe Bayern, goaudio’s clients also include “BILD News Update,” “MOZ Briefing,” and “Aponet in 3 Minutes.” The startup sees itself as an AI infrastructure for German-language audio production—with a focus that clearly sets it apart from generic text-to-speech services: Fairly and securely licensed voice clones of professional voice actors, GDPR-compliant operations, and a workflow that doesn’t replace editorial style but makes it scalable.