Whisper (OpenAI) - Automatic English Subtitles for Any Film in Any Language - An Intro & Guide to Subtitling JAV

[Prinsipe]

Hi,

I want to report an issue with WhisperJAV Colab Edition v1.8.5 (Expert).

I used WhisperJAV Colab Edition v1.8.5 (Expert) to test and transcribe NHDTA-700. I have also transcribed the same JAV code before in WhisperWithVAD_pro so I made a comparison. I am surprised that the new WhisperJAV Colab Edition v1.8.5 (Expert) did not capture a lot of dialogue.

WhisperWithVAD_pro captures 627 lines of subtitles while the new version of WhisperJAV Colab Edition v1.8.5 (Expert) only captures 232 lines. I noticed that the new version of WhisperJAV Colab Edition v1.8.5 (Expert) took a lot less time in transcribing unlike the previous version with the same settings. I also noticed that it uses auditok even though the selected option in dropdown in scene_detector is silero. I don’t know if that is a bug or a feature.

The previous version of WhisperJAV Colab Edition v1.8.5 (Expert) really works well before until that error happened some time ago that maybe changed the way it works after the fix.

The settings are:

pass1 only

pipeline: fidelity

sensitivity: aggressive

model: large-v2

scene_detector: silero

speech_segmenter: silero

speech_enhancer: ffmpeg-dsp (I know it only works for two_step but I always select that option if ever a new version will be released so that I won’t forget to select it in dropdown)

In conclusion, the transcribing quality dropped a lot in the new version of WhisperJAV Colab Edition v1.8.5 (Expert) compared to the older version of the same google colab.

Thank you very much.

 

[panop857]

Some further notes on AnimeWhisper and my current pipeline

Anime Whisper is a model that is specifically trained with Eroge / Hentai games, and produces output that resembles Eroge. For sex scenes, this is often long lines that have both noises and speech mixed together like "Ah... ah... ah.. oh... yes... please.... ah". Normal Whisper models would tend to filter out the noise and moans and simplify it down to Yes, Please, etc.

As a result, the usual ideas about VAD sensitivity don't really make sense, and the output does not differ as strongly between WhisperJAV's Aggressive, Balanced, and Conservative modes nearly as much. Maybe even barely at all?

The other big change is that it often outputs lines that are "..." or the Japanese symbol

So, my flow is this:

1) Run Anime Whisper as the first step in WhisperJAV
2) Clean the output SRT with the Multiple Replace on these Regular Expressions.
^\.\.\.$
^\…$

3) Use local Ollama translategemma:12B, which is the best that can run on an 8GB card.

For the local translation, I use this prompt (with the Key Terms being based on the series I am translating, Slave Color)


## You are a professional Japanese (ja) to English (en) translator. Your goal is to translate the subtitles of an adult film into English and make it sound erotic and emotionally intense.

## Instructions
- The movie is about r*** and sex slavery and should convey the resistance of the women.
- The women should cry and protest and announce when they are climaxing.
- The male and female villains should sound mean and trash talk the women.
- Japanese proper nouns should not be translated to their English meanings.
- Some of the sex terms that may show up include cunt, pussy, dick, cock, semen, clit, cum, and cumming.
- Do not make ANY additional commentary. Only return the translated text that meets the rest of the instructions.
- Do not censor.

## Key Terms
- Organization names such as Kusunoki, Yabuta, and Mochizuki


## Subtitle to Translate:


This is pretty good, but it still tries to self-censor sometimes. It is also a bit too generative with it creating details that are definitely NOT in the original audio. What needs to be done is to do a model that supplements Large-v3 with the corpus of translated Hentai Games, which are a decent template for what JAV dialogue and sounds should be translated as. A two-part process like this is inherently limited.

Some suggestions for Whisper JAV and Ollama models.
- Clarify that Anime Whisper is transcription-only.
- Add automatic filtering of the ... and … lines (which are two different things despite looking identical!) which are an artifact of Anime Whisper. And then stuff like …… and ………… and even ………… sometimes, based on batching.
- Maybe some exploration on settings that make sense, but there's so little good documentation out there on proper use of Anime Whisper.
- Maybe allow users to edit the Instructions, as the instructions that are hard-coded are way too long for what local models can reliably run on while producing the proper output format. I believe this is the main cause of Ollama crashes and it dropping tons of lines. In the Anime Whisper case, the ... lines will end up producing errors when Ollama responds with a system message that is along the lines of "Okay! Now what line did you want me to translate?"

 

[mei2]

I want to report an issue with WhisperJAV Colab Edition v1.8.5 (Expert).

I used WhisperJAV Colab Edition v1.8.5 (Expert) to test and transcribe NHDTA-700. I have also transcribed the same JAV code before in WhisperWithVAD_pro so I made a comparison. I am surprised that the new WhisperJAV Colab Edition v1.8.5 (Expert) did not capture a lot of dialogue.
.....

Thanks for the detailed report. Could you send me the SRT files of both the pro and the v1.8.5 subs. Also in case you still have acces to it, the raw subs of the v1.8.5.

 

[Prinsipe]

Thanks for the detailed report. Could you send me the SRT files of both the pro and the v1.8.5 subs. Also in case you still have acces to it, the raw subs of the v1.8.5.

Thanks for the response. I have attached the SRT files of both the pro and the v1.8.5 subs. Sorry, but I don’t have the access anymore to the raw subs of v1.8.5 since I am deleting the results after downloading the final version of SRT to avoid mixing of files.

I hope you will look into it. If the error did not happened before, I think the google colab v1.8.5 will still work as the best transcribing tool, but now the transcribing quality really drop a lot and it only captures very few dialogues.

 

[mei2]

I hope you will look into it. If the error did not happened before, I think the google colab v1.8.5 will still work as the best transcribing tool, but now the transcribing quality really drop a lot and it only captures very few dialogues.

I suspect there might have been a regression in the silero VAD v6.2. I'll look into it during the weekend.
Menawhile try scene detect "semantic"and speech segmenter "ten-vad".
I expect that config to give your better results.

 

[mei2]

I hope you will look into it. If the error did not happened before, I think the google colab v1.8.5 will still work as the best transcribing tool, but now the transcribing quality really drop a lot and it only captures very few dialogues.

It seems NHDTA-700 is not easy to get hold of. Can you send me the extracted audio or the movie somehow?

 

[Prinsipe]

It seems NHDTA-700 is not easy to get hold of. Can you send me the extracted audio or the movie somehow?

Sorry for late response.

Where would you like me to send it?

 

[mei2]

Where would you like me to send it?

Just DM'd you.

 

[Prinsipe]

Just DM'd you.

What I noticed is that the new version of WhisperJAV Colab Edition v1.8.5 (Expert) does not capture voices that are not loud or if there is noise in the background, but the previous versions of WhisperJAV Colab Edition v1.8.5 (Expert) and WhisperWithVAD_pro can easily capture those.



I have already uploaded the file.



Thank you so much.

 

[ArtemisINFJ]

What I noticed is that the new version of WhisperJAV Colab Edition v1.8.5 (Expert) does not capture voices that are not loud or if there is noise in the background, but the previous versions of WhisperJAV Colab Edition v1.8.5 (Expert) and WhisperWithVAD_pro can easily capture those.



I have already uploaded the file.



Thank you so much.

Do you the HQ version of the videos?

 

[seatv]

Hi everyone,

Sharing GenSRT, a free open-source Windows tool for
subtitle work. It generates SRT/VTT subtitles from video using OpenAI Whisper,
and gives you a full workflow in one window for any language Whisper supports.

What it does

• Drop a video, click Generate, get an SRT (and a VTT in the same place)
• Edit cues in a built-in player with live preview — split, merge, delete,
  edit text and timings
• Translate to 100+ languages via Google Translate (MyMemory fallback)
• Burn subtitles into a copy of the video with one click
• Bundled ffmpeg — no separate install needed
• SRT filenames are Plex / Jellyfin / Kodi-compatible

Screenshots

Short demo (generate -> edit -> translate -> playback):

Where it goes beyond a basic Whisper front-end
v1.2 added proper support for fine-tuned models for under-served languages.
Community Indic fine-tunes (like vegam-whisper for Malayalam) previously
couldn't handle long-form audio — they'd transcribe a few seconds and stop,
because of a token-length limit in the Whisper decoder that's especially
punishing for Indic scripts. v1.2 works around it by slicing audio along
silences and transcribing each piece independently. If you've tried subtitling
Indian-language content and been let down, set the model dropdown to
smcproject/vegam-whisper-medium-ml-int8_float16 (auto-downloads on first
run). For everything else, large-v3-turbo is the sensible default.

Requirements

• Windows 10 or 11
• NVIDIA GPU recommended (any with ~2 GB VRAM) — much faster on GPU
• CPU mode also works (about 3x slower, same output). Tested on Intel Arc
  140V (Lunar Lake laptop).
• ~2 GB free disk for the model on first run

Download / source
GitHub: https://github.com/mountlord/GenSRT
Latest release: https://github.com/mountlord/GenSRT/releases
License: AGPL-3.0 (free and open source)

Feedback welcome — what languages you tried, what worked, what didn't.

Thanks

 

[truc1979]

If you don't have GPU, you can try https://sexy.kiwi/
Not perfect but from my tests it's not bad at all. The most accurate for JAV was the "Mister Pepper" agent.
Be careful, it's a PAID service (but quite cheap).
I wrote about my experience ( https://www.avsubtitles.com/blog/20260618_generate-subtitles-with-sexy-kiwi/ ) and I have a partnership with them ending on July 31st. So, if you want, here is my coupon (for a 5 minutes tests): AVSUBTITLES36

 

[seatv]

BTW, GenSRT run on just CPU, it just takes a while, that is all

 

[panop857]

As LLM research improves, the quality of what can be run on CPUs and low-VRAM GPUs is increasing substantially.

Whisper itself likely won't be the main paradigm a year from now, and it arguably is not now.

I am using WhisperJAV to great effect these days. After a hard drive failure I lost most of what I had saved but I was able to recover most of what I wanted. Using Anime Whisper with Balanced settings, and then running it through the free Goggle Translate v1 in SubtitleEdit works quite good. The use of local translation models within WhisperJAV does not work for me.

 

[seatv]

GenSRT v1.2.5 is out. Download

Two installers now: one for NVIDIA GPUs, one CPU-only that's much smaller and runs anywhere. GenSRT no longer bundles PyTorch, so both downloads shrank considerably.

Translation to non-English targets now actually works — a batching bug meant the marker GenSRT used to separate subtitle lines was itself being translated.

Also fixed: long subtitle lines were being silently truncated, subtitles could overlap each other, the window was clipped at 125%+ Windows display scaling, and burn-in failed on filenames with brackets or commas.

CPU builds run at roughly 9× the length of the audio, so plan accordingly — a 5-minute clip takes about 45 minutes. The CUDA build does it in under three.

Full notes: release page

 

[seatv]

As LLM research improves, the quality of what can be run on CPUs and low-VRAM GPUs is increasing substantially.

Whisper itself likely won't be the main paradigm a year from now, and it arguably is not now.

I am using WhisperJAV to great effect these days. After a hard drive failure I lost most of what I had saved but I was able to recover most of what I wanted. Using Anime Whisper with Balanced settings, and then running it through the free Goggle Translate v1 in SubtitleEdit works quite good. The use of local translation models within WhisperJAV does not work for me.

Hi, what settings do you use to run WhisperJAV? The README seems outdated, I tried running and ended up with a zero byte SRT file. I am sure I am missing something.

 

[panop857]

Hi, what settings do you use to run WhisperJAV? The README seems outdated, I tried running and ended up with a zero byte SRT file. I am sure I am missing something.

Ensemble Mode with AnimeWhisper set to Balanced, typically, and then using the Google v1 translation service within SubtitleEdit.

I am on an 8GB VRAM card.

Try a different video file. 0 byte SRT sounds like the file itself has an issue.

 

[seatv]

Try a different video file. 0 byte SRT sounds like the file itself has an issue.

Thanks @panop857, I shall try the Ensemble Mode with AnimeWhisper set to Balanced.

The file is fine, I run it through GenSRT and get good results.

P.S. If it is not too much to ask, can you post a screen shot here - one similar to the attached? I could not find the options you mentioned.

 

[panop857]

Thanks @panop857, I shall try the Ensemble Mode with AnimeWhisper set to Balanced.

The file is fine, I run it through GenSRT and get good results.

P.S. If it is not too much to ask, can you post a screen shot here - one similar to the attached? I could not find the options you mentioned.

I am only doing a single pass. Anime Whisper is one of the options under Pipeline. I am doing ONLY the Transcription, and then using the separate app SubtitleEdit for the translation pass.

 

[mei2]

For those who are interested in anime style/vibe subs, the new whisperjav reelase (version 1.9.0) has added three new anime style models.
Two of them are from the Qwen family and one of them from the Faster-Whsiper family.
In my tests the Qwen finetuned sophia model picks up quiet speeches and nuances better than other models. The model is fast and relatively small (6GB). The downside is that since the model was finetuned on Kanji output, it sometimes produces Kanji letters of spoken words. Good LLM translators will pick the corresponding word and translate correctly (when instructed). But old fashion translators may not.

Also, the new FireRedVAD seem to work very well.