Wrong speech recognition on RTX 4060 ti

[zavalroman]

When using RTX 4060 ti 16G for speech recognition in a long audio file, the result is incorrect with many repeating sentences. However, when switching to RTX 3060 12G, the result is excellent. I'm trying to recognize Russian speech (parameter -l ru), but the problem occurs with both Russian and English audio files.

On video example using the audio file: https://commons.wikimedia.org/wiki/File:060123-John.Willinsky-The.Economics.of.Knowledge.as.a.Public.Good.ogg

Video example with reproduce log: https://youtu.be/oWXSz7V7sZA

full_reproduce_log.txt