Vers. 0.4 beta is out! Major update for noScribe

[kaixxx]

Version 0.4 is a major update, so many things can go wrong...

• Please post your experiences here (positive or negative).
• You can also open an issue to report an error.

Thank you!

[tobias-d]

First of all: Thank you for the update and the software which has profoundly improved my working life.

Unfortunately, the recent Beta crashes after loading the whisper plugin. The speaker recognition works but than the app crashes completely. I am on Windows 11.

[kaixxx]

Regarding CPU usage, @gernophil and @BabyFnord: I would be happy to add such an option in the advanced-options. But I am not sure how to implement it. With faster-whisper, it is easy: there is an option to reduce the number of cores used. But for pyannote, I don't know how to do this. Any ideas?

[kaixxx]

@Schmukas95 @tobias-d @amru39
I've released a new version 0.4.1 which should fix the issue. Please test and report back if you'll find the time, thanks!

This bug was really annoying me because it came in only a few hours after I made vers. 0.4 non-beta. Thank you for your help in narrowing the problem down!
I am thinking about reintroducing GPU-support in the future since it could bring down transcription times by a lot. Unfortunatly, I don't have a suitable graphics card (NVIDIA only). Would you help testing such a new version? It may take a few weeks until I get around working on this.

[amru39]

It seems that we have found the error. Can you confirm that it runs fine with the exe linked above? I'm about to update the installer. Thanks!

Yes, it worked. Attaching log file for reference.
KII6.log

[amru39]

@Schmukas95 @tobias-d @amru39 I've released a new version 0.4.1 which should fix the issue. Please test and report back if you'll find the time, thanks!

This bug was really annoying me because it came in only a few hours after I made vers. 0.4 non-beta. Thank you for your help in narrowing the problem down! I am thinking about reintroducing GPU-support in the future since it could bring down transcription times by a lot. Unfortunatly, I don't have a suitable graphics card (NVIDIA only). Would you help testing such a new version? It may take a few weeks until I get around working on this.

Testing it right now. Will report back later today/tomorrow. Yes, I can help to test any further versions. I have a couple of files that I need to transcribe.

[amru39]

Quick update. Worked perfectly. With no other background process/application running, the same recording took 15 minutes less to transcribe. The CPU usage during embedding was around 50 percent and for the actual transcription around 80 percent.

[kaixxx]

I have compiled a new version of noScribe. This is not going to fix the original problem, but should give us more information in the log files to narrow down the error.
You can download the new noScribe.exe from here: https://drive.switch.ch/index.php/s/dmnjyNGI16XRQ4a This is not an installer, just the main application.
You have to place the file inside the installation folder of noScribe 0.4, which should be "C:\Program Files (x86)\noScribe". Please replace the exe that is already there (you can also rename the old exe if you want to keep it).
If you now run noScribe.exe, a terminal window will appear which will give some information during the transcrption process. More importantly, the log file should now actually have some info in it.
Could you post the logfile here? Please delete any sensitive information beforehand.
Thanks!

[kaixxx]

One addition: You can test with a much shorter section than 15 minutes to make your life easier. This should not make a difference. Just set the "Stop" field to 00:01:00 for the first minute of your audio.

[tobias-d]

Okay, it is working now. Thanks a lot! I am also happy to test with my NVIDIA card, would be very curious how fast it can be with a RTX A4000 :)

[kaixxx]

@tobias-d Perfect, happy to hear that. I'll come back to you if I find the time to work on the cuda-integration.

[tobias-d]

Won’t be able to test it before Monday. But yes, also have a powerful NVIDIA Graphics Card.

[kaixxx]

Thank you. Let's see - I hope you'll have a working version by monday.

[BabyFnord]

On a side-note, the macOS version provided by @gernophil runs 💯 fine on my late-2012 MacBook Pro and macOS Mojave 10.14.6. You'll need to exchange two .dylibs (see attached .zip), though. To anyone inclined on using trailing edge Macs & macOS for noScribe, the following Terminal commands get the adaptation done on v0.4 beta (download and unpack that .zip first):
cp $HOME/Downloads/libtiff.6.dylib /Applications/noScribe.app/Contents/Frameworks/PIL/__dot__dylibs/
cp $HOME/Downloads/libX11.6.dylib /Applications/noScribe.app/Contents/Frameworks/av/__dot__dylibs/
noScribe_x64_Libs.zip

Thanks to @kaixxx and @gernophil for their patient beta-test support!

[kaixxx]

Note: Vers. 0.4 had a problem in combination with NVIDIA graphics cards on Windows. Vers. 0.4.1 fixes this. If you did not experience this bug, you don't have to update.

[gernophil]

Just realeased 0.4.2 for macOS, which is using torchaudio 2.1.0. If you are using MPS (macOS GPU) then I strongly recommend to switch to that Version, since the old speaker recognition was really error prone.

[kaixxx]

@Schmukas95 @tobias-d @amru39
A new version with cuda is out. Could you help testing it? See #50

[tobias-d]

Sorry for the late reply. I have been using the new version quite a lot recently and it works perfectly fine. Transcription times vary of course but it is very fast.
Last one: 1hr recording with 3 speakers: 100min

I was wondering if the new version also has an effect on the quality of the transcription. I had the feeling it was more accurate but it is hard to tell because there are so many factors to consider.

[kaixxx]

Great, thank you. The transcription quality should be the same, but its hard to test really. However, there will be a new version released in the next couple of days with some small improvements. Watch out..