I have 2 machines – one running Fedora 36 and the other running Windows 10. Both machines have kdenlive 22.08.1 installed. I want to be able to generate subtitles, so both machines have the vosk-model-en-us-daanzu-20200905 and vosk-model-en-us-0.22 speech models installed.

On the Fedora machine I was able to use the vosk-model-en-us-daanzu-20200905 speech model to generate subtitles when the machine had 8GB of memory, but the larger speech model caused the program to crash (no reason given). I increased the memory to 16GB and the larger speech model now performs fine. I did run System Monitor before and after the memory change and I could see that, with the large speech model, the memory did get completely utilised when the memory was too small.

On the Windows machine with either speech model it all appears to work: It starts by ‘Starting audio export’ then moves on to ‘Starting speech recognition’ which sits there for a while on 0% and then it says ‘Subtitles imported’. However, watching Task Manager, in particular, Memory usage, it hasn’t done much and, indeed, the subtitles are completely missing. Originally, that machine had 4GB of memory and I increased it to 8GB (the maximum according to the motherboard specs) but that made no difference.

It would appear that on the Windows machine, it is not attempting to load the speech model – is that correct? What can I do about it?

I had similar issues. Since then, I use the “small” language pack available. Try a smaller language pack.