Release Subtitle Edit 4.0.2

[niksedk]

It's time for the next update - 4.0.2 :)

This version includes layouts (which replaces Show/hide video and Show/hide waveform). The new layouts make it much better to create subtitles for a mobile video in 9:16 format and/or using a vertical monitor. Layouts also make it possible to have video left or a few other options (layouts have shortcuts too):

This version also includes:

• Rewritten auto-translate with new translators.
• Remove interjections are now language specific, easier to edit, and have a skip list.
• Setup a language filter if you only use a few languages.
• A few minor fixes for the new controls introduced in SE 4.
• (Regression) SE now again remembers font size for text box.

Lots of minor improvements + bug fixes: https://raw.githubusercontent.com/SubtitleEdit/subtitleedit/master/Changelog.txt

Give the beta a test run ❤️ https://github.com/SubtitleEdit/subtitleedit/releases/download/4.0.1/SubtitleEditBeta.zip

[uckthis]

Unbreak Text shortcut no longer works in texbox. Still works in list.

[Purfview]

Released Whisper-Faster r160.6. :) r160.5 has autodownload bug. [not affecting usage in SE]

EDIT: I think there is a bug in large-v3 configs, so a new version could be out soon. :D

[JDTR75]

EDIT: I think there is a bug in large-v3 configs, so a new version could be out soon. :D

You mean the original large-v3 model? I was doing some "tests" yesterday and got a hallucination error with faster-whisper-large-v3-fp16

[Purfview]

You mean the original large-v3 model?

What is "original large-3 model"?

I was doing some "tests" yesterday and got a hallucination error with faster-whisper-large-v3-fp16.

Not getting hallucinations with "large-v3" would indicate that something is wrong. 😛

[JDTR75]

By "original" I mean the one from OpenAI-Whisper (Python). From my limited understanding, different implementations need the models to be converted - correct me if I'm wrong 😅. So when you mention bugs in the configs, you mean for your conversion or in the "original" OpenAI-Whisper?

Not getting hallucinations with "large-v3" would indicate that something is wrong. 😛

Didn't get hallucinations with large-v2 or medium for the same file, just fyi.

[Purfview]

In converted model. Anyway, I didn't noticed any influence on results with fixed config.

Didn't get hallucinations with large-v2 or medium for the same file, just fyi.

That's expected.

[despairTK]

SE 测试版使用 Purfview 的 Faster Whisper 进行了大型 v3 更新：https://github.com/SubtitleEdit/subtitleedit/releases/download/4.0.1/SubtitleEditBeta.zip

https://github.com/ggerganov/whisper.cpp/releases/tag/v1.5.0 whisper.cpp is currently updated to version 1.5, which supports large-v3 and Distil Whisper models, supports GPU operation, and has some other improvements and new features.

[niksedk]

https://github.com/ggerganov/whisper.cpp/releases/tag/v1.5.0 whisper.cpp is currently updated to version 1.5, which supports large-v3 and Distil Whisper models, supports GPU operation, and has some other improvements and new features.

Latest beta now includes Whisper CPP 1.5.0: https://github.com/SubtitleEdit/subtitleedit/releases/download/4.0.1/SubtitleEditBeta.zip Let me know how it works.

[uckthis]

Latest beta now includes Whisper CPP 1.5.0: https://github.com/SubtitleEdit/subtitleedit/releases/download/4.0.1/SubtitleEditBeta.zip Let me know how it works.

It's super, super fast. Transcribed a 22 minutes file in 2.5 minutes with Large v1 model. I tried Large-q5 (V3) and it took 15 minutes on the same file and the whole file was !!!!!!!!!!!!. Also, unlike Faster, it saves file automatically and you have to open the SRT in SE to edit it, it doesn't automatically open in SE.

[JDTR75]

@uckthis you got this with CPU?

[uckthis]

@JDTR75 24GB GPU, but it used less than 8GB.

[JDTR75]

Dumb question... how did you get it working with the GPU? Directly from SE? I was trying but it's going for the CPU 😓

[uckthis]

@JDTR75 You need Cuda 12. Lesser won't work. Directly from SE, yes.

[niksedk]

https://developer.nvidia.com/cuda-downloads

[uckthis]

Tested Large V2 as well, and it is as fast as V1 too. Not sure if V3 will work or if it's worth the effort. OpenAI now has tasted GPT money, they won't put resources towards an open source model.

I still get random periods when Post Processing is checked, though it's a lot better than before. Had only three instances in today's 22-minute file.

[niksedk]

I still get random periods when Post Processing is checked, though it's a lot better than before. Had only three instances in today's 22-minute file.

OK, could you show these errors?

[uckthis]

Video.zip Check with Start with uppercase letter after paragraph... 3 instances. whisper_log.txt

[JDTR75]

https://developer.nvidia.com/cuda-downloads

Oh! Thank you so much @niksedk. Working fine now :)

[darnn]

The download for the large model in ConstMe is now broken. And I also got a file full of 30-second chunks of "!!!!!" with CPPcublas v3. Should installing CUDA solve this? Because I thought I had installed it already.

[uckthis]

The download for the large model in ConstMe is now broken.

I think Large is now Large V3 in Const. They changed some shit.

[Martyprod]

Just downloaded last beta, and tried Model V3 with purfiew faster whisper (without doing anything else). option :compute -int8 (1050Ti graphic card). it output a 7 bytes file totally empty. it was doing the same before until the option of -int8 was added to SubtitleEdit. testing V2 and seems to work as usual.

[JDTR75]

Noticed a bug when trying to use Whisper for specific segments. It's not letting me change/choose the Whisper engine I want. It's default on one.

[niksedk]

Noticed a bug when trying to use Whisper for specific segments. It's not letting me change/choose the Whisper engine I want. It's default on one.

Seems to work here... could you supply a bit more info?

[Purfview]

Released Faster-Whisper r160.7. :)

[niksedk]

Released Faster-Whisper r160.7. :)

Beta updated: https://github.com/SubtitleEdit/subtitleedit/releases/download/4.0.1/SubtitleEditBeta.zip

[uckthis]

I see that CPP cuBLAS use Cuda 12 and Faster uses Cuda 11. Could this be the reason why CPP is faster than Faster? CPP is almost twice as fast, and I don't see any significant difference in quality of output.

Also, random periods got even better this time. Only two in a 44-minute file.

[niksedk]

Also, random periods got even better this time. Only two in a 44-minute file.

Could you add the srt file?

[uckthis]

Could you add the srt file?

1.txt

Funnily enough, I was testing a random file and didn't save it, so I had to redo it and still the errors remain.

[JDTR75]

Noticed a bug when trying to use Whisper for specific segments. It's not letting me change/choose the Whisper engine I want. It's default on one.

Seems to work here... could you supply a bit more info?

Let me show you @niksedk. I'm selecting CPP cuBLAS but it's going with Faster-Whisper. This is on beta 443.

https://github.com/SubtitleEdit/subtitleedit/assets/1891941/a38338d4-7e92-4212-b671-6bb9c9998106

UPDATE: It's not happening in the latest beta 459 :).

[Purfview]

I see that CPP cuBLAS use Cuda 12 and Faster uses Cuda 11. Could this be the reason why CPP is faster than Faster? CPP is almost twice as fast, and I don't see any significant difference in quality of output.

@uckthis Because Faster-Whisper is tuned for quality. Can you share "txt" and "srt" files made by CPP "large-3" on this file ->https://we.tl/t-3I4JanmGK3 ? Don't enable post-processing.

[uckthis]

@Purfview Large V3 isn't working on CPP. It only produces !!!!!!!!!!!!!!

[Purfview]

Medium then.

[JDTR75]

I see that CPP cuBLAS use Cuda 12 and Faster uses Cuda 11. Could this be the reason why CPP is faster than Faster? CPP is almost twice as fast, and I don't see any significant difference in quality of output.

Also, random periods got even better this time. Only two in a 44-minute file.

Interesting. Faster-Whisper is still faster on my end, just fyi.

[uckthis]

@Purfview 1.txt

V2 Large

[Purfview]

@Purfview 1.txt

Yeap, CPP didn't output accurate timestamps.

[uckthis]

@Purfview Any significant difference in transcription quality?

[Purfview]

Any significant difference in transcription quality?

No.

[Nomad234]

https://developer.nvidia.com/cuda-downloads

novice question here: I'm currently running on a gtx1060. Is this cuda download compatible with my gpu? and is downloading it going to mess with stable diffusion automatic1111 or comfyui on my pc?

[Nomad234]

https://developer.nvidia.com/cuda-downloads

I've updated to newest cuda. My tests are done on mainly Arabic voice recordings that have many English terms and words within the narrative. Some remarks that might be helpful:

1. CPP works on CPU only, but detection of English words is much much better than Faster Whisper (though most of the detected English words are transliterated in Arabic).
2. CPP with CuBLAS always returns an error that cudart64_12.dll is missing. However, when I open the whisper log file, I find that the voice audio has been already transcribed, albeit with a lot of jargon detonating the progression of the transcription process. It worth noting however, that transcription here is blazing fast. For comparison sake: transcription of a 5 minute sample with CPP on my 8th gen i7 (16 gb ram) took almost 17 minutes. On CPP with CuBLASS (with the missing dll error), it took mere 4 seconds !!!!

All CPP tests were done with large-q5_0 model.

I hope there is a way to fix this dll error and have a proper srt format within STE. Anyways, great effort. Thanks a million!

[uckthis]

@Nomad234 Install Cuda 12.3 from here. File is 3GB. https://developer.nvidia.com/cuda-downloads

[Nomad234]

@Nomad234 Install Cuda 12.3 from here. File is 3GB. https://developer.nvidia.com/cuda-downloads

I did. Twice. Still, the same error message "cudart64_12.dll is missing"

[diomed]

why did you release 4.0.2 without merging last additions for croatian ocr? :-/

[niksedk]

why did you release 4.0.2 without merging last additions for croatian ocr? :-/

Sorry, I forgot - merged now.

[niksedk]

@Nomad234 Install Cuda 12.3 from here. File is 3GB. https://developer.nvidia.com/cuda-downloads

I did. Twice. Still, the same error message "cudart64_12.dll is missing"

Is the dll file on your machine and in path?

[niksedk]

Subtitle Edit 4.0.2: https://github.com/SubtitleEdit/subtitleedit/releases

[uckthis]

I did. Twice. Still, the same error message "cudart64_12.dll is missing"

Use your head, bro. Search cudart64_12.dll in the Cuda installed directory and paste it in CPPcuBLAS folder. If the error persists, try pasting it in different locations where CPP might be accessing it from.

[Nomad234]

I'm a 55 year old bloke here who's has no background in tech. Sorry if I annoyed any one by my dumb question, but you could just scroll over. Even better, you could have given your advice without condescending. And of course I searched the cudart64_12.dll in my system, couldn't find it. Even downloaded the dll file from a dll website, and put it in the CPP cuBLAS folder, and it still didn't work. Owing to my lack of knowledge, I don't know of what other locations CPP might access. Perhaps, it just has to do with my old gtx 1060, and it's not compatible with some aspects of CUDA 12.

[uckthis]

@Nomad234 cudart64_12.zip I'm dumber than you when it comes to tech. I didn't mean to come across as condescending, it's just how we talk these days.

Anyway, I'm attaching the file from my system, hopefully it works. It's highly possible your GPU is too old to use new Cuda.

[Nomad234]

@Nomad234 cudart64_12.zip I'm dumber than you when it comes to tech. I didn't mean to come across as condescending, it's just how we talk these days.

Anyway, I'm attaching the file from my system, hopefully it works. It's highly possible your GPU is too old to use new Cuda.

Unfortunately, didn't work out. But thanks a lot for helping me out. Appreciate it!

[diomed]

I dont see where I can change color of selected subtitle line, even tho I remember you working on it and merging it.