On 2/10/26 3:21 PM, Zbigniew Jędrzejewski-Szmek wrote:
On Tue, Feb 10, 2026 at 08:16:08PM +0000, Zbigniew Jędrzejewski-Szmek wrote:* AGREED: APPROVED (+5, 1, -0) (@conan_kudo:matrix.org, 18:39:01)The text that was voted: Change is approved. FESCo recommends that at least one small useful model is packaged.
In my experience, Whisper's smallest models don't work very well. Their website even says that the tiny model has low accuracy, and the small model is slow.
I would recommend packaging the base model. It worked well on my few-years-old 4-core laptop, transcribing at 6x the real time, with impressive accuracy.
One thing I could not get whisper to do was segmentation: it could not simultaneously differentiate speakers and transcribe. This made it less useful for podcasts, sadly.
-- _______________________________________________ devel mailing list -- [email protected] To unsubscribe send an email to [email protected] Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/[email protected] Do not reply to spam, report it: https://forge.fedoraproject.org/infra/tickets/issues/new
