Nao GLP-1
- Thread starter Besh
- Start date
-
Akiba-Online is sponsored by FileJoker.
FileJoker is a required filehost for all new posts and content replies in the Direct Downloads subforums.
Failure to include FileJoker links for Direct Download posts will result in deletion of your posts or worse.
For more information see this thread.
@mei2
On a different note, I DL'ed your program instead of using the google colab. It is day and night, The speed is faster, no limit on how many I can do in one day, I have no idea why I havent done that before.
On a Core I7 12700k , RTX 4080 16gb and 64gb RAM, I did 17 in less than 2 hours on Balance/agressive direct to english. Pass it through LADA after to remove pixel and boy, I dont think I can go back to ''normal''.
-Besh
On a different note, I DL'ed your program instead of using the google colab. It is day and night, The speed is faster, no limit on how many I can do in one day, I have no idea why I havent done that before.
On a Core I7 12700k , RTX 4080 16gb and 64gb RAM, I did 17 in less than 2 hours on Balance/agressive direct to english. Pass it through LADA after to remove pixel and boy, I dont think I can go back to ''normal''.
-Besh
I always wondered why someone want tu sub a Japanese speaking movie with japanese subtitles? I mean, I wont sub an english movie since I understand english, what am I missing here?
-Besh
There are 2 things at play
I started writing this and it became way too long
I think there are few things at play:
1- A distinction between transcription, vs translation. For an English movie, the captions (subtitles) that we see are in fact transcriptions. You and I don't need to translate the transcription because we undersatnd the original language. Similarly for Japanese movies, transcriptions are in the original language. Those can then be translated to target languages "faithfully". For any non-English target languages, it is likely to achieve higher quality (higher fidelity) subtitles if they are translated from the source language. For example, most of WhisperJAV users are in China. The reslut of Chinese subs are better if they have been translated from Japanese than from English.
2- Many LLMs, and few ASRs (like Whisper) are able to output direct to target langauge captions. That very much depends on their training. For example Whisper is trained on Japanese clips with English subtitles. It is also trained for Japanese clips with Japanese subtitles. Whisper models are trained much longer on Japanese subtitles and English sutitles. Therefore Whsiper shows much better WER and CER rates for direct transcribed Japanese. With the current tech stack, the quality of Japanese transcriptions are higher than direct to English.
3- This will change in future as most of the subtitling trend is to use LLMs. The reason that it is not happening yet, is cost and censorship. Most of the powerful LLMs block NSFW and JAV content at the moment.
I started writing this and it became way too long
I think there are few things at play:
1- A distinction between transcription, vs translation. For an English movie, the captions (subtitles) that we see are in fact transcriptions. You and I don't need to translate the transcription because we undersatnd the original language. Similarly for Japanese movies, transcriptions are in the original language. Those can then be translated to target languages "faithfully". For any non-English target languages, it is likely to achieve higher quality (higher fidelity) subtitles if they are translated from the source language. For example, most of WhisperJAV users are in China. The reslut of Chinese subs are better if they have been translated from Japanese than from English.
2- Many LLMs, and few ASRs (like Whisper) are able to output direct to target langauge captions. That very much depends on their training. For example Whisper is trained on Japanese clips with English subtitles. It is also trained for Japanese clips with Japanese subtitles. Whisper models are trained much longer on Japanese subtitles and English sutitles. Therefore Whsiper shows much better WER and CER rates for direct transcribed Japanese. With the current tech stack, the quality of Japanese transcriptions are higher than direct to English.
3- This will change in future as most of the subtitling trend is to use LLMs. The reason that it is not happening yet, is cost and censorship. Most of the powerful LLMs block NSFW and JAV content at the moment.
