AI Speech-to-text is poor and generates off timestamp.

Emma-Eriksson wrote on 5/7/2025, 9:28 AM

How come when I use speech-to-text, the AI analyses it very poorly in swedish?

And how come the genereted subtitle ends up on wrong location (even tho I generated it from Source Media, not the timeline)

The end-result often results in the subtitle starting waaay too early and then maybe stops when we stop talking in the video?

Br,

Emma

Comments

RogerS wrote on 5/7/2025, 10:31 AM

Could be a limitation of the underlying Microsoft Azure service- the online Swedish language dataset maybe isn't so big compared to English or other languages it does a better job on.