Models that decide fast and models that think slow
Fast models, reasoning models and the new family of models that only make decisions, like Jev: what each brings to a voice agent, where latency rules, and how to combine them.
Research
We research how people talk to machines. And we build what we learn.
Since 2017 we've been researching applied AI for voice, conversation and multimodal agents, with companies, universities and public institutions.
Research areas
Spanish speech synthesis
Our own text-to-speech models, brand voices, Spanish accents and co-official languages. It started with a CDTI NEOTEC project and continues in Fonos.
Synthetic voice and disinformation
Synthetic voices to train and evaluate fake audio detection systems, with the RTVE-UGR Chair.
Diversity in voice
Voices that represent more people, like Free the Voices, the first LGBTIQ+ synthetic voice bank, created with LLYC.
Conversation design and evaluation
Assistant and agent personality, conversational quality and user-centered evaluation methods.
Accessible information
Voice and AI to help information reach more people, like RTVE's election coverage in small towns.
Multimodal agents
Agents that take on tasks across voice, text and images, while people make the calls that matter. It builds on our voice-and-screen experiences, awarded at the 2019 Alexa Skills Challenge: Multimodal.
Projects and collaborations
RTVE-UGR Chair
A research chair run by RTVE, Spain's public broadcaster, and the University of Granada. Synthetic voices to research fake audio detection. Scientific paper published in 2024.
RTVE: elections and weather
Synthetic voices for local election and weather news, in Spanish and Catalan.
NEOTEC · CDTI
Research on our own Spanish speech synthesis models, the origin of Fonos.
Universities
Collaborations with Universidad Rey Juan Carlos, Universidad de Granada, Universidad de Castilla-La Mancha and Universitat de Lleida.
Want to turn your research into a prototype?
We research and collaborate with
From the blog
Articles on voice, conversation and AI.
What would this face sound like? Synthetic voices from a single photo
At IberSpeech 2026 we presented a method that generates a plausible voice from a photo. Here's how it works, what we measured and how far it goes.
Building a chatbot in 2018 and building one today
From Alexa intents and flows to agents built on language models: what has changed in building a chatbot or voice assistant, and what hasn't.
Work with us
We join R&D projects, research chairs and funding calls as a technology partner. If you have a research line in speech, language or conversation, let's talk.