Mati Staniszewski
speaker
2,430 appearances
3 recordings
1 series
first heard Sep 2026
last heard 6 Sep
Mati Staniszewski’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 3.
Appearances
So we added then speech to text model, transcription model.
So you speak at analysis and you have the text generated in a in a quick way.
Then we added localization model that brings from one language to another, and you can still have the same voice.
Then we add
It orchestration model or interaction model for voice agents.
So you can have speech to text, the LLM, the text to speech, and you can effectively speak with the agent and it will seamlessly orchestrate when to pause, when to answer, how you can interact it.
So that was kind of our research journey over last years.
And you know, it's um uh so far we've been able to lead against the biggest, biggest
Labs, biggest hyperscalers, on quality, on benchmarks and and win against them.
Yeah, the at the end of you know it it started even smaller, right?
It was like uh we were we started as a nine million dollar company and then before that was zero of course.
So it's uh um but I think that the that speaks to the the genius of the research team where
You didn't need scale of resources.
It wasn't compute problem.
It wasn't data problem.
It was the architecture problem.
Can you create the novel architecture of how the models can translate the text into speech in a new way?
How you encode and decode audio, how you understand the context, how you understand and decode and encode voices.
Um so Piotr uh was able to assemble some of the best researchers in the world to to to to to
To actually fix that.
Showing 321–340 of 2,430 · page 17 of 122
← Previous
Next →