Zoom ASR captioning

[VeroZac]

Hi!

I am a university student and interpreter. I am doing research on the accuracy of Zoom autogenerated captions - how they can help interpreters during the task, possible issues (like foreign names, abbreviations etc.) and such.

I have produced an experimental speech that will be delivered by native speakers in 3 languages (English, Spanish and Czech). It contains high number of aforementioned units.

But, while most of the speech is as realistic as possible, some sentences were added just for the sake of adding more foreign names or numbers. They are still authentic, but might feel a little bit "out of place" for human listener. Zoom, at least to my knowledge, has never released anything talking about whether their ASR system uses Natural Language Processing or anything that would help me understand whether or not this might impact the accuracy of the output.

It is impossible to contact Zoom support unless I'm a paying customer and even at that, not with this kind of query. Anyone has any knowledge or links to point me towards?

[colegs]

@VeroZac - In the Automated Captions support article, it does highlight some limitations, including specifically calling out that it is not recommended for compliance or accessibility purposes.

[VeroZac]

Thank you, I know about that, the research's purpose is to have a look at the accuracy, not to prove their captions are amazing. But I do want to provide best conditions possible. I am aware of the possible limitations they mention, so there will be no background noise, no interrupting speakers etc. I am just curious if the accuracy might be impacted if the experimental speech is not 100% authentic (because it is still experimental). In case the captions were using something like Natural Language Processing that looks for context and meaning, it would be better to have completely authentic speech.

[colegs]

Sorry, I can't provide any insights into the Zoom logic for doing transcriptions.