[DigitalToday reporter Seulgi Son] AI language model company DeepL said on March 26 it is ahead of the built-in translation caption features of major collaboration platforms such as Google Meet, Microsoft Teams and Zoom in both quality and stability, citing results from an independent benchmark on its voice translation solution, DeepL Voice.

The survey was conducted by translation and language AI research firm Slator at DeepL’s request. It ran blind evaluations across 14 language pairs, including Korean and English, and 96 percent of language experts selected DeepL Voice as their top choice.

Translation quality scores were 96.4 for DeepL Voice for Zoom and 96.3 for DeepL Voice for Teams. That was above the competing platforms’ average of 87 to 89 points. The rate of major translation errors was 76 percent lower on average than other platforms, and the share of translated segments meeting the pass standard was 79 percent, about double the competitors’ average of 42 percent.

Caption stability scores were 88.6 for DeepL Voice for Zoom and 85.8 for DeepL Voice for Teams. On-screen caption changes and flickering fell 37.6 percent versus Microsoft Teams and 54.7 percent versus Zoom. Slator analysed on-screen rendered captions frame by frame to measure flickering, shaking and the frequency of edits.

Jarek Kutylowski (야렉 쿠틸로브스키), DeepL’s founder and CEO, said, "Language AI is core infrastructure for operating global companies," and added, "Accuracy and stability are not optional but essential."

DeepL signalled a global update on April 16. It is expected to unveil improvements to DeepL Voice and an expansion of cross-platform collaboration features.