Blackmagic Forum

Hi,
I successfully created a voice model of my own voice from around 20 minutes of material of me. I was then trying to use voice convert on a 1 minute clip of my colleague. But every time I try it, no matter if I select render in place or new track, the process starts and after a few seconds Resolve crashes to desktop. Using the built-in voice models also crashes to desktop.

Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
Windows 11 Pro, 24H2
Geforce RTX 2080

Diagnostics file is attached.

Heiko, just a very wild guess - but in your logs I've seen you are using folder names like this:
E:\!WORK\
E:\!JOBS\!Davinci Resolve Backups\

Shouldn't be a problem with Windows per se as its an allowed character, but maybe it does trip over some of the AI tools inside of Resolve?

Hi Robert, thanks for the reply. The funny thing is, I was trying to demo the crash to a friend of mine yesterday during a zoom call and the voice effect worked. One time. It did replace the voice but right as was trying to change the voice on another clip in the same timeline, I had the crash to desktop again and after that it crashed everytime just like before. Absolutely no clue what was different this one time where it worked. Did not change anything in the settings or on my system between tests.

As most of the AI stuff is using pytorch in some ways, maybe it creates the temp files but then can't re-read them later or crashes due to the "!" in the path names.

Or maybe it's a Nvidia driver bug - who knows. Your driver version is slightly newer than the minimum required for DVR20.

Thank you for taking the time. I followed your advice and removed any references to any of the folders with the "!" in the name. I went so far to create a fresh database on the C drive with no special characters in the path name. Changed the backup location for Resolve to that folder too. I moved the files I was trying the feature with from my network drive to a locally attached USB-C SSD and imported them into a fresh project from there. All other folder references in Resolve point to a folder named cache in the root of the C drive.

Still the same result, immediate crash to desktop after a few seconds when the voice change process starts and the progress bar shows. Also no matter if I use my created voice model or the vanilla ones that come with R20, crash every time.

I'm having the exact same issue, with crashing to the desktop in Win 11. When this happens, Davinci stays in memory, and the process cannot be stopped without rebooting. Initially, I was able to convert audio successfully using both the built-in voices and one that I created. But then it began crashing when I would try to convert, and it hasn't worked since. I even tried restoring my C drive (on which Davinci is installed) from a backup made before I installed version 20, then reinstalling Davinci, but the crashes continued.

I'm experiencing the same issue.
Trying one of the built-in models.
It immediately crashes, without prompting for a crash report.

I have tried this on two machines. On the first - 14700K with nVidia 3060 - it does a really good job on some clips and on other it crashes when analysing the clips. On the second (9900K with 2080 super) it created the voice model but whenever I try to do a convert the slider just sits there and nothing happens.

I assumed this was just a problem as it is beta. On the clips on which it did work I was very impressed.

Voice convert randomly crashes to desktop. Happens using one of the four built in voice models or one that I created. Certain audio clips work fine, others crash. Tried to pin down the differences between the clips that work and the ones that don't. Occasionally if I take a clip that crashes and split it in two, it works. No real consistency though. Clearing the CacheClip folder between crashes doesn't have any effect. Tried the latest NVidia studio and game ready drivers. No difference. Not using any folder names with "!" in the name.

Intel(R) Core(TM) i7-13700K CPU @ 3.40GHz
Windows 11 Pro, 24H2
Geforce RTX 3060
NVidia 572.83

My Voice Convert crashes Davinci every single time unless I turn off "Use neural engine optimization on NVIDIA".


UPDATE - DOWNLOADED THE LATEST STUDIO DRIVER AND IT NOW WORKS BETTER, BUT SOMETIMES CRASHES

Turning off the Nvidia neural engines worked for me as well! Thanks for figuring this out.

Just installed DaVinci Resolve 20 Public Beta 2. Voice convert still crashes to desktop most of the time. But the good news is, I tried what Chardiot suggested and turned off "Use neural engine optimization on NVIDIA" and that solves the problem. Since you don't really use the voice convert feature that often, I guess it's not that big of a deal to turn off neural optimization when doing a voice convert. Hopefully they will fix it in later updates, but Public Beta 2 still has the bug.

Intel(R) Core(TM) i7-13700K CPU @ 3.40GHz
Windows 11 Pro, 24H2
Geforce RTX 3060
NVidia 576.02

Can confirm too, Beta 2 and no neural engine optimization on NVIDIA also solved the crashes on my machine.

I, too, can confirm that disabling the NVIDIA Neural Engine solved my AI Voice Convert crashes. Thank you for the suggestion.

Just installed DaVinci Resolve 20 Public Beta 3. So far, no crashes to desktop when doing voice convert with "Use neural engine optimization on NVIDIA" turned on. Did about 10 voice converts as a test. Thank you Blackmagic ...

Intel(R) Core(TM) i7-13700K CPU @ 3.40GHz
Windows 11 Pro, 24H2
Geforce RTX 3060 12GB
NVidia 576.02

I don't know if it still a thing, however - for me the this problem was fixed by generating optimized media. No crashes even with Nvidia ai optimization on.