I just used dialog transcription for the first time, was actually just on a recording of people talking to get the text not actual dialog for a movie and it seemed to work well. However I decided to compare it against Spectralayers since that also has a similar function. I haven’t checked accuracy yet but the noticeable things was speed. Spectralayers used my GPU and cranked out the hour long file in a little over a minute. Nuendo took about 30 minutes on Balanced mode on the CPU, and it wasn’t using the whole CPU only a couple threads.
Now I realize the models are likely different, but if the Nuendo model could be made to use a GPU, if available, it would be nice. It didn’t look like it needed that much memory, maybe 4GB, so most GPUs should actually have enough VRAM to run it.