It should be working now. If any errors please dm me the error outputs and also which version you'r eusing. The latest version is 1.8.0.
akiba resident JAV subtitlers & subtitle talk★NOT A SUB REQUEST THREAD★
- Thread starter desioner
- 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.
It should be working now. If any errors please dm me the error outputs and also which version you'r eusing. The latest version is 1.8.0.
Google Colab
colab.research.google.com
Use this one, its updated, might have fixed the error you mentioned
Colab Error
Any help appreciated
Error Message on colab Step 2
The error ModuleNotFoundError: No module named 'librosa' means that
a required audio processing library is missing from the WhisperJAV installation.
I will add a step to install librosa into the WhisperJAV environment to fix this.
SystemExit: Transcription failed
Thx jasial for the new colab url
Thx mei2 for updating colab
Any help appreciated
Error Message on colab Step 2
The error ModuleNotFoundError: No module named 'librosa' means that
a required audio processing library is missing from the WhisperJAV installation.
I will add a step to install librosa into the WhisperJAV environment to fix this.
SystemExit: Transcription failed
Thx jasial for the new colab url
Thx mei2 for updating colab
Quick note:
I have updated the expert editions. It is now using 1.8.2 version.
I have also updated the whisperwithvad_pro version to fix the issue.
They all should work but I haven't had bandwidth to test them. Hope no more issue
Let me know
I'm sure I saw a recent post here about the new Qwen3-ASR model. I would have replied to that post but now I can't find it.
I had a play with the two sizes of the Qwen model and (with a quick first try) it doesn't seem that great at transcription compared to Whisper Large v2. I'm hoping it's just the settings I'm using that give me the underwhelming results and it can actually do better than my first attempt.
The zip attached has transcribed and translated subtitles for this interview (Actress Japan: Nozomi Ishihara) on YouTube. I will give the models another go with some longer JAV-sourced audio when I get time.
I had a play with the two sizes of the Qwen model and (with a quick first try) it doesn't seem that great at transcription compared to Whisper Large v2. I'm hoping it's just the settings I'm using that give me the underwhelming results and it can actually do better than my first attempt.
The zip attached has transcribed and translated subtitles for this interview (Actress Japan: Nozomi Ishihara) on YouTube. I will give the models another go with some longer JAV-sourced audio when I get time.
What's your opinion on:
tencent/Hunyuan-MT-7B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
TranslateGemma - a google Collection
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Gemma 2 JPN Release - a google Collection
A Gemma 2 2B model fine-tuned on Japanese text. It supports the Japanese language the same level of performance of EN only queries on Gemma 2.
huggingface.co
webbigdata/gemma-2-2b-jpn-it-translate · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
GitHub - litagin02/anime-whisper
Contribute to litagin02/anime-whisper development by creating an account on GitHub.
github.com
I haven’t tried any of these models.What's your opinion on:
tencent/Hunyuan-MT-7B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.coTranslateGemma - a google Collection
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.coGemma 2 JPN Release - a google Collection
A Gemma 2 2B model fine-tuned on Japanese text. It supports the Japanese language the same level of performance of EN only queries on Gemma 2.huggingface.cowebbigdata/gemma-2-2b-jpn-it-translate · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.co
GitHub - litagin02/anime-whisper
Contribute to litagin02/anime-whisper development by creating an account on GitHub.
For text translation, without censorship issues, I still haven’t found anything better than Deepseek - and there’s usually a slight improvement with each new version.
If I get some time (time always seems in short supply) I might try the fine tuned Whisper models you’ve listed for transcription.
I built a local-only pipeline that used anime-whisper and was quite happy with how it performed. I don't speak Japanese, so i can't verify the transcriptions directly, but the translated output had more explicit language compared to the generic whisper models, or even the kotota-tech japanese specialty model.
I might dust that project off, but I pivoted to using WhisperJAV rather than continuing to try and vibe-code a better translation pipeline.
Qwen3-ASR looks very interesting
I might dust that project off, but I pivoted to using WhisperJAV rather than continuing to try and vibe-code a better translation pipeline.
Qwen3-ASR looks very interesting
for noobs/ inexperienced people like myself, whats a good tutorial on how to use these?What's your opinion on:
tencent/Hunyuan-MT-7B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.coTranslateGemma - a google Collection
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.coGemma 2 JPN Release - a google Collection
A Gemma 2 2B model fine-tuned on Japanese text. It supports the Japanese language the same level of performance of EN only queries on Gemma 2.huggingface.cowebbigdata/gemma-2-2b-jpn-it-translate · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.coGitHub - litagin02/anime-whisper
Contribute to litagin02/anime-whisper development by creating an account on GitHub.
Here's an odd question:
Has anyone tried subtitling a video and the time-codes just don't seem to line up, despite tweaking?
I had a video, created subs for it, and the timing was just off in certain places. Trying to adjust the timings in AegisSub, but it seemed to just cause other issues.
I found another copy of the video, tried again-similar issues?
Then I went to SubtitleCat, found someone *else's* sub for it...it was also off.
So...are there some videos that just mess with the timings? Maybe how they are encoded?
Anyways, it's just one video, just an odd occurrence.
Has anyone tried subtitling a video and the time-codes just don't seem to line up, despite tweaking?
I had a video, created subs for it, and the timing was just off in certain places. Trying to adjust the timings in AegisSub, but it seemed to just cause other issues.
I found another copy of the video, tried again-similar issues?
Then I went to SubtitleCat, found someone *else's* sub for it...it was also off.
So...are there some videos that just mess with the timings? Maybe how they are encoded?
Anyways, it's just one video, just an odd occurrence.
Last edited:
Yes, I've had that happen a few times over the years.
I see that you found a different copy, but....was it a different version?
Meaning, was it a different size, length, etc?
I had some success finding different versions, sometimes bigger is better...sometimes smaller.
I could never find any other solution.
Good luck.
If the audio has corruption, it'll cause subs to not line up anymore since it'll skip that part when playing it. It might not be noticeable when playing it if that happens over silence.
You should still be able to change the timings to match, but just doing it globally may not be enough since it'll be fine before the corruption and not after so it would need to be done in at least 2 chunks.
Could also be a playback issue if your pc has a hard time playing that video, it might drop frames and the subs will slowly get out of sync.
That would be nearly impossible to fix the timing for.
Also, some people can edit the video slightly when ripping it like removing studio intros, adding ads, merging parts and cutting transitions in the process and many more possibilities.
That does happen a lot less these days, but it was much more common on older videos and depending on what's been done, it can make timing it more troublesome.
Also, you can't rely on subtitle edit or aegisub internal video player to time the subtitles since they use directshow codecs to load it which are not frame accurate(but can load pretty much any video, which is why they use it) and that will get worse the more you seek in there so there will be a difference between what you see in there and when you play it in another player to watch it.
You should still be able to change the timings to match, but just doing it globally may not be enough since it'll be fine before the corruption and not after so it would need to be done in at least 2 chunks.
Could also be a playback issue if your pc has a hard time playing that video, it might drop frames and the subs will slowly get out of sync.
That would be nearly impossible to fix the timing for.
Also, some people can edit the video slightly when ripping it like removing studio intros, adding ads, merging parts and cutting transitions in the process and many more possibilities.
That does happen a lot less these days, but it was much more common on older videos and depending on what's been done, it can make timing it more troublesome.
Also, you can't rely on subtitle edit or aegisub internal video player to time the subtitles since they use directshow codecs to load it which are not frame accurate(but can load pretty much any video, which is why they use it) and that will get worse the more you seek in there so there will be a difference between what you see in there and when you play it in another player to watch it.
In my experience, webrip versions that use variable rates (VBR / VFR) cause the rift. The rift is between extarcted audio vs source video. I.e. any subtitle that is made based on the audio, will rift and does not stay in sync with the video.
If the rip is not corrupted, give a try to this command:
Code:
ffmpeg -fflags +genpts \
-i javmovie.mp4 \
-map 0:a:0 \
-ac 1 \
-ar 16000 \
-c:a pcm_s16le \
-af "aresample=resampler=soxr:precision=28" \
-avoid_negative_ts make_zero \
javmovie.wavOther more intrusive command (it alters the sync points) if the first one didn't work:
Code:
ffmpeg -fflags +genpts \
-i javmovie.np4 \
-map 0:a:0 \
-ac 1 \
-ar 16000 \
-c:a pcm_s16le \
-af "aresample=resampler=soxr:precision=28:async=1:first_pts=0:min_hard_comp=0.1" \
-avoid_negative_ts make_zero \
javmovie.wavIf these don't work and you cannot find a good source to redownload, then best would be to remux/re-encode the entire movie to be connstant bit rate / frame rate.
Last edited:
