Most voice AI is built for English and clean audio, then adapted for everyone else. We're building it the other way round. Half a billion people speak Spanish, in accents from Bogotá and Mexico City to Buenos Aires and Madrid. Their voice AI should understand how they actually talk.
Simulated calls in Colombian and European Spanish. Grey words are partial hypotheses; highlighted words are recognised slots.
Streaming speech recognition for calls, built for 8 kHz audio and the details a call turns on.
A full-duplex model that listens while it speaks, handles interruptions and hands tasks to your agents mid-conversation.
Agents built end to end on our own models.
Latency is what the caller feels, so we design for it first.
Trained on Spanish across Latin America and Spain, not translated from English.
Our models are public. Our API runs them in production for you.
Hosted in the EU or inside your network. Call audio stays under your control.
We're onboarding our first pilot partners now.
Streaming over WebSocket or SIP, priced per minute. Hosted in the EU under a data processing agreement.
For banks and call centres that can't send audio outside their own network.
The model adapted to your callers' accents and your product vocabulary.
Ablero is a voice AI lab founded by Igor Kuzmin, an NLP researcher formerly in the Language Technologies team at the Barcelona Supercomputing Center and a PhD researcher at Pompeu Fabra University.
femto-asr on Hugging Face →Real-time Spanish voice AI for call centres, BPOs and voice-agent teams across Latin America and Spain.