adverse-sapphire•2mo ago
Generating "Spoken Cantonese" Subtitles with Subtitle Generator?
Hi,
I'm wondering if there are any plans to have an option for Spoken Cantonese subtitles to be generated with the "Generate Subtitles" option, as for now (when it works) it appears to only generate written Chinese.
Thanks!
Jarrad
10 Replies
You mean instead of generating simplified/traditional mandarin to generate ones for Cantonese?
adverse-sapphireOP•2mo ago
Yeah. Although some videos I see on YT say "Cantonese - HK" for subtitles but they are still Written Chinese. Looking for a 口語 option. Not 書面語.
unwilling-turquoise•2mo ago
This would be a whisper limitation.
Not sure what you mean by written Chinese exactly, or you mean the subtitles aren’t a one to one of what they’re saying (so a transliteration instead I think those are called).
But also yeah it depends if whisper supports Cantonese or not, from my understanding Cantonese support for things like this usually aren’t available or work well @HulK
adverse-sapphireOP•2mo ago
Exactly. Cantonese, unlike Mandarin, has a different written form vs spoken. The vast majority of the time the subtitles in Cantonese videos will be different than what the speaker is actually saying because they are using the written form, which has been this way for a long time. Since I'm not a native speaker, my goal is to mostly learn how to speak proficiently and the added layer of not being able to use most subtitles is a challenge. That being said, I know there's plenty of great work being done on other discord channels to create more spoken word subtitles for content, but in this case I was just curious about Migaku's subtitle generator/transcription abilities.
Given that Noah just pointed out it uses Whisper, I can start looking around and see what I find. Thanks to both of you for your responses!
Oooh that’s interesting so Cantonese written and spoken sound like they differ a lot how interesting, I’m learning simplified mandarin so I was confused at first lol but that explains why finding Cantonese support for things tends to be harder from my understanding. But yeah they use whisper so I’d look into if they have any support or are planning support for it.
optimistic-gold•2mo ago
@Jarrad - Tiles from the Crypt Yes there are plans
It's not a whisper limitation.
adverse-sapphireOP•2mo ago
That's great to hear
unwilling-turquoise•2mo ago
My bad. I was assuming on what I heard here and should you have articulated that as an assumption more clearly.
optimistic-gold•2mo ago
The issue is that default Whisper doesn't have good enough Cantonese output, but I think I found a model that will work pretty well. Just has to make it up the prio list