At Prisa Media, we have always believed that trust is the foundation of everything we do.

For decades, our brands have built a deep bond with audiences across Spain and Latin America, especially through our spoken-word radio stations — Cadena SER, Caracol Radio, ADN, and W Radio — which shape the news agenda in Spain, Colombia, Chile, and Mexico.

Cadena SER in Spain and Caracol Radio in Colombia have been reporting the news for over 100 years and 75 years, respectively, and are part of their countries’ sonic history. 

Voice, of course, lies at the heart of our work. That’s why, three years ago, we began exploring the potential of AI in this field and embarked on creating our own synthetic brand voice, Victoria, the voice of soccer, which was awarded Best Audio Product at the 2023 INMA Global Media Awards.

Since then, we have witnessed firsthand the dizzying pace of AI, especially in its application to voice cloning. We realised the technology in this field was advancing so rapidly that each day it became harder to distinguish authentic audio from synthetic fakes.

Prisa Media is the largest producer of Spanish-language audio in the world, and as such, we knew we couldn’t stand on the sidelines. We felt an urgent responsibility to act.

Fighting against fake audio

Thanks to the support of Google and the technical expertise of Plaiground, the AI division of the Spanish technology firm Minsait, we got to work.

In addition to conceptualising how to integrate AI technology into our regular fact-checking processes, it was crucial to explore which architectures would deliver the best results in such an emerging and technically demanding field as voice cloning.

From the outset, VerificAudio, which won first place in the national brands Best Use of AI in Customer-Facing Products category at the INMA Global Media Awards in May, had two core goals:

• First, we wanted to empower journalists by embedding AI into the fact-checking process, enabling them to analyse suspicious audio quickly and reliably.
• Second, we aimed to engage the public by creating a direct bridge between listeners and newsrooms, inviting citizens to submit audio files they found suspicious for expert verification.

In December 2023, after months of research, VerificAudio was launched as a proof of concept — just as global elections were ramping up and audio deepfakes were entering the public arena in dangerous ways: fake messages from political leaders, false endorsements from journalists, and viral scams flooding social media and WhatsApp, especially in Latin American countries.

How we built it

Building VerificAudio wasn’t just a technical project; it was a shared cross-border effort. The project was initially conceived in Colombia, with the teams at W Radio and Caracol Radio, and soon evolved into a transversal initiative reaching our radio stations in Spain, Mexico, and Chile.

On the technological front, we built two custom AI models: a comparison model that checks suspicious audio against a verified recording, and an identifier model that examines the physical attributes of audio to detect synthetic speech.

We dedicated 2024 to refining and retraining both models on a robust dataset of hundreds of hours of real and synthetic audio.

Equally important was making these tools useful and accessible to our newsrooms. We worked side by side with our journalists, gathered feedback, and devoted 2024 to refining our audio fact-checking tool, ensuring it worked seamlessly — from laptops in Madrid to mobile phones in Bogotá.

They needed a tool that was fast, intuitive, and mobile-friendly. As a result, we developed a fully responsive Web app, integrated it into newsroom workflows, and even added features like video-to-audio separation and a searchable log of past verifications.

On the public-facing side, we launched VerificAudio.ai, a multi-country Web platform that allows media outlets to submit suspicious audio clips. Using geolocation, the site routes users to country-specific pages, where they can learn about local disinformation cases, access resources on how to spot deepfakes, and contribute directly to the verification process.

Gaining attention

The response has been deeply motivating. Since launching, we’ve received verification requests from across the Spanish-speaking world and beyond — including from major international players like Reuters, Univisión, and The Poynter Institute, as well as prestigious national media outlets like Chequeado, La Silla Vacía, and Efecto Cocuyo, among others.

Media outlets from Venezuela to Mexico and from Argentina to Spain have submitted audio clips ranging from political hoaxes to health scams.

In the past year, we have analysed more than 70 suspicious audio files. We’ve tackled everything from fake presidential statements in Venezuela to deceptive endorsements by celebrities and journalists in Mexico.

One of the most rewarding aspects, however, has been the project’s impact within our own newsrooms. We’ve trained cross-country verification teams, created a shared repository of analysed cases, and developed new editorial protocols.

From the inside, we’ve learned that the challenge of audio deepfakes is as cultural as it is technological. People don’t just need tools — they need to understand what deepfakes are, how they spread, and why they matter.

That’s why one of our main goals for the coming months is to ensure that VerificAudio becomes not just a verification service, but also an educational platform.

Looking ahead, we’re also aiming to expand into other minority languages, explore audio watermarking and video verification, and continue refining our audio fact-checking models to keep pace with the latest developments in the field.

At Prisa Media, we are driven by the conviction that trust is worth protecting — and that it’s something we must defend together. We believe that journalism has a crucial role to play in this, but so do our audiences.

VerificAudio is also our invitation to them to join us in this fight.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .