Piotr Dabkowski

Dąbkowski studied for an MPhil at Cambridge. His 2017 saliency paper with Yarin Gal is described below.[7][4]

ElevenLabs identifies Dąbkowski as its CTO and co-founder and records that he previously worked at Google. His GitHub profile also identifies his role in ElevenLabs research. In a jointly authored company update, the founders describe expanding from speech generation into transcription, sound effects, music, dubbing and conversation. These are company research programs; the announcement does not assign him sole authorship of the underlying models.[2][1][3]

Explaining image classifiers

Dąbkowski co-authored Real Time Image Saliency for Black Box Classifiers with Yarin Gal in 2017. The method trains a model to mask parts of an image and observe how a classifier’s scores change. The resulting map points to regions important to the classification. After training, it produces a map in one forward pass, making the approach suitable for rapid inspection. The paper evaluates image datasets; its explanation method should not be read as proof that a classifier reasons like a person.[4]

JavaScript inside Python

His Js2Py repository contains a JavaScript interpreter and translator implemented in Python. It makes code from one programming environment usable from another.[5]

The Eleven v4 announcement, co-authored with Staniszewski, frames speech generation as more than reading words aloud. The same words can need different emotion, pacing and delivery depending on a scene or conversation. It presents a model designed to respond to that context while maintaining speaker identity, alongside a Turbo variant for applications sensitive to response time. This is the founders’ explanation of the release, with demonstrations and company-reported evaluations.[6]