Cutoff for final character is not correctly timed in Chinese transcriptions - How to fix?

[rookes]

Thanks to Purf for this tool. It really has helped a ton on some of my community Chinese transcription projects.

I have a recurring issue when transcribing Cantonese, but I assume it is present in Mandarin Chinese as well. For basically every single line, the final character gets cut off early. I am using the Faster_Whisper_XXL_Pro version "--realign" option to improve timings, but it doesn't seem to help. Very consistently, the subtitle lines will start a bit before the audio correctly, and then will cut off the moment the final character is spoken (not when the final character ends).

Here's an example from One Punch Man:

As you can see, this is an extreme example where he shouts the last character for a long time. But it is present with normal speech as well. Here's a more normal example.

I don't know if this is a quirk with Whisper in general and I'm not sure if there is a solution. I currently just run some scripts to add time onto every subtitle, but it is very inconsistent. Is there a command line option or some other potential change that could fix this issue and make each line end at the end of the actual speech?

[Purfview]

Where do you find this in "Cantonese"? I've looked at nyaa, there is only Japanese.

[rookes]

The Cantonese dub used to be available on Viu but I can't find it there anymore. Let me see if I can find it and I will message you so you can test it out

[rookes]

Sent to you via email