Incorrect, your video shows correct behavior here but not ripple delete. Maybe I did not test that contrived use case. Confirmed, it works with between first and second of 3 clips but not just 2 clips.
Contrivedā¦ I didnāt know that word 
Synonyms found: Artificial, Invented, Improvised, Complicated etcā¦
Can you give me an example of an uncontrived use case when using ripple delete?
It works with between first and second of 3 clips but not just 2 clips. I just fixed that. It was not working if transition included the last clip on the timeline.
1 Like
Did you try a different video?
I do not know why whisper.cpp-main.exe is failing on your machine. Maybe it is using some CPU instruction not available on your system.
Yes, I did try a few videos. I also tried with a few different audio clips (WAV and MP3) on their own audio track.
What is the model of your CPU? How much RAM do you have?
This is just a guess, but maybe the problem is that your processor does not support AVX2 instructions. I am not sure if the speech-to-text engine can detect if those instructions are missing.
I donāt know if this is a reliable website, but itās all I could find for now.
They mention AVX, but not AVX2
Have you checked this bug in the official beta version, and not in ānightlyā?
Yes, I did. Same problem.
I did a pretty crazy experiment, I ran the beta version of shotcut in the Winlator emulator on my Android smartphone, and speech-to-text recognition works fine on it. I donāt think my Snapdragon 8gen2 has any support for AVX commands, but it works, so the problem is not in the processor.
I do not consider an emulator or CPU translator a valid testāat least not realistic. But I do not know how to audit a binary executableās usage of illegal instructions. Often now, they use runtime CPU detection, and that sounds like it can run into bugs considering the number of CPUs there have been over time. For what itās worth, I tested a couple of other older machines where I have Windows 10 installed, and they worked. The oldest I have is:
- Processor Intel(R) Coreā¢ i5-4460 CPU @ 3.20GHz, 3201 Mhz, 4 Core(s), 4 Logical Processor(s)
I am confident this is the issue now. I reviewed the whisper.cpp code, and it does not appear to use runtime CPU feature detectionāonly build-time and the features can be overridden. I checked our build, and it is building with AVX2 and FMA. Plus this reported whisper.cpp issue.
We could disable AVX2; however, that would be a very big sacrifice for something already slow on CPU. I will consider to make 2 builds with and without AVX2, but I would need to figure out a way to detect and choose the correct one. That could come after the release.
@MusicalBox was this a quick test or are you actually editing videos on an 11year old cpu? Is itā¦ bearable? I have a 3 years old intel laptop and if Iām on battery it takes at least a full second just to update the UI after clicking a video in timeline. Itās almost usable if I enable high performance mode or plug it in but thank god for modern ryzen desktop cpus or Iād never try video editing at all.
I do have a more recent computer (a laptop) in my shop but it is not convenient or comfortable to do editing over there. So yes, I do use that old 11-year-old computer to edit videos. But I donāt find it unbearable. Iām not experiencing this āfull second just to update the UI after clicking a video in timelineā you talk about.
I very rarely create 4K and/or 60fps videos. I divide long projects in shorter (about 10 to 15 minutes) sub-projects, and I always try to find ways to use as few tracks and filters as I can. Most of the time I will convert video files to edit-friendly. I find that it makes editing (scrubbing in particular) a lot smoother. And of course, Proxy and Preview Scaling helps a lot.
So yeah, that PC is old, but itās still adequate for what I do. And, for now, I donāt feel the need to invest (waste) $1000 to $3000 on a new machine.
3 Likes
Just for extra information, I Googled āhas my computer got AVX2ā and it gave quite a lot of helpful information which I thought may be of interest:
1 Like
Well, I asked Gemini, and this answer might also help users working on a prehistorical computer (like I do apparently 
) :
1 Like
By the way @shotcut, thanks for the fix on removing transitions with Ripple Delete. Works fine now 
2 Likes
I recognize both that this particular code is someone elseās and that the alphabet soup of optional CPU features makes it a nightmare to produce programs that take advantage of whatever features may be present in a userās CPU - and especially so for a program that targets multiple architectures.
Having said that, to the extent itās reasonable to do so, it would be nice to have runtime detection and utilization of processor features (and Iām speaking in general, not specifically about AI speech-to-text). Video editing is resource-intensive, and anything that can use a feature like AVX2 to improve performance on a CPU that supports it would certainly be nice to have. Obviously, much lower priority than fixing bugs or adding important features! But definitely nice to have.