WhisperAttack - OpenAI Whisper Voice Recognition and injection into VoiceAttack

[nikoel]

59 minutes ago, markturner1960 said:

Thanks, trying to install this now....stuck at FFmpeg...I dont see a link for thuis? How is it installed please? I looked in the unzipped Whisperattack folder but was unsure what to do from there...

Hey mate. We are about to release an executable,

However if you want to get it going straight away put the following into the command prompt run as admin:

winget install --id=ffmpeg.ffmpeg -e

 

[markturner1960]

Thanks mate, I have got to here and its stopped......

[sleighzy]

5 hours ago, markturner1960 said:

Thanks mate, I have got to here and its stopped......

That’s fine, just leave that Window open and you’ll see the transcriptions of your recorded voice logged there. 

[markturner1960]

OK, so can I assume its now properly installed? I am at the stage where I have a folder called "WhisperAttack v0.4 beta" but I cant see a folder anywhere called "whisperattack server command" which I am supposed to put in voice attack apps folder......

I have a folder called voiceattack plug in, in my WhisperAttack v0.4 beta folder.....which has a "voiceattack server command .dll" in it.......but I am not seeing what is described in the instructions......sorry! 

[sleighzy]

On 2/10/2025 at 9:45 AM, markturner1960 said:

OK, so can I assume its now properly installed? I am at the stage where I have a folder called "WhisperAttack v0.4 beta" but I cant see a folder anywhere called "whisperattack server command" which I am supposed to put in voice attack apps folder......

I have a folder called voiceattack plug in, in my WhisperAttack v0.4 beta folder.....which has a "voiceattack server command .dll" in it.......but I am not seeing what is described in the instructions......sorry! 

I double-checked the release now and there is a folder in there named VoiceAttack Plugin you can rename that folder to WhisperAttackServerCommand to match the README instructions before putting it in the VoiceAttack Apps folder. We can rename that in a subsequent release to match the docs.

You can jump into the VR4DCS Discord server as well. The #vaicom-voiceattack channel in there is where we're hanging out so can get some more timely troubleshooting feedback.

https://discord.com/channels/610534461456777257/809527129422430218

 

Edited April 21 by sleighzy

[markturner1960]

Cheers, will do that....

[sleighzy]

There's an executable version of this available here. https://drive.google.com/drive/folders/1z-a-dgxeQS6Aj_jNq2mhuZNBfFwbaoNP?usp=sharing

This does not require you to install Python or ffmpeg.

Download and unzip the file. Put the contents in the folder of your choice, e.g. C:\Program Files\WhisperAttack, just make sure they're all together. You can create a shortcut to the WhisperAttack.exe file.

When you run this it will run in the background and put an icon in your system tray. You can right-click on that to exit it. Closing VoiceAttack will also exit it automatically.

Logs will get written to C:\Users\username\AppData\Local\WhisperAttack\WhisperAttack.log , you may need to open and close the log file if your text editor does not auto-refresh it.

First load may take a little bit as it downloads the Whisper model.

Let us know how you get on with it and if you encounter any issues. This executable application will be our way going forward in the near future.

null

[KIWIvolshebnik]

I'm trying to get this working using the executable, but I don't think WhisperAttack and VoiceAttack are communicating?
I've attached a screenshot of VoiceAttack and the log from WhisperAttack. 

I've checked that I've followed the installation and set up instructions correctly. I edited the settings file to point to the VoiceAttack installation:

voiceattack_location=F:\Program Files\VoiceAttack\VoiceAttack.exe

The log file shows:

'voiceattack_location': 'F:\\Program Files\\VoiceAttack\\VoiceAttack.exe'

Not sure if that is just a formatting thing in the log file (the double \)

WhisperAttack.log

[nikoel]

Hey Team

I've also ran into issues with the new executable. It's still under development

However, I have just pushed an update that addresses the issues that some users have reported. For instance, when user double clicks the file the window closes itself. I also updated the VA Profile so it now runs the plugin by default

Please let me know if it works (or doesn't) - for now the executable works, but in some cases doesn't. Sleighzy will have a fix in two shakes of a lamb's tail. However here is the new version which runs the script as it did originally

https://github.com/nikoelt/WhisperAttack/releases/tag/v0.4.1-beta

All the best

 

[sleighzy]

I've uploaded a new version of the v0.4.1-beta executable. You will need the full contents when unzipping it. It can be found in the same location, https://drive.google.com/drive/folders/1z-a-dgxeQS6Aj_jNq2mhuZNBfFwbaoNP?usp=sharing

This still has the system tray icon, but also has a console window (closing this will also stop WhisperAttack) so you can the model being loaded and any fatal errors. All other logging, e.g. transcribed text, will still go to the log file and not the console window.

Edited February 11 by sleighzy

[KIWIvolshebnik]

I've downloaded the v0.4.1-beta and it seems to be working (kinda).

I get the following error in the log file:

ERROR - Failed to transcribe audio: Library cublas64_12.dll is not found or cannot be loaded

 

[sleighzy]

1 hour ago, KIWIvolshebnik said:

I've downloaded the v0.4.1-beta and it seems to be working (kinda).

I get the following error in the log file:

ERROR - Failed to transcribe audio: Library cublas64_12.dll is not found or cannot be loaded

 

Hmmm, sounds vaguely familiar, let me double-check. Can you attach your full log please. Is this the first time using it, i.e. you hadn't tried the previous version (which I've removed from that Google drive but still have on hand).

And to confirm, you haven't run any of the previous Python versions of this previously, i.e. the non-executable ones? Or is this the Python one?

Edited February 11 by sleighzy

[KIWIvolshebnik]

57 minutes ago, sleighzy said:

Hmmm, sounds vaguely familiar, let me double-check. Can you attach your full log please. Is this the first time using it, i.e. you hadn't tried the previous version (which I've removed from that Google drive but still have on hand).

And to confirm, you haven't run any of the previous Python versions of this previously, i.e. the non-executable ones? Or is this the Python one?

Full log attached.
I tried both the python version and the previous executable version.
This is from the most recent executable version you posted above.

WhisperAttack.log

[sleighzy]

11 minutes ago, KIWIvolshebnik said:

Full log attached.
I tried both the python version and the previous executable version.
This is from the most recent executable version you posted above.

WhisperAttack.log 2.67 kB · 0 downloads

Can you run the below commands, run each of them separately, agree to any messages which ask you if you wish to remove packages. It's fine if it doesn't locate a package, this is expected, but let me know which ones it doesn't find. The last command will then install a different version of torch (it's possible in one of the earlier versions of WhisperAttack that it installed an incorrect version. Then try both the executable and the Python version to see if this makes a difference.

pip uninstall torch

pip uninstall whisper

pip uninstall openai-whisper

pip install torch --index-url https://download.pytorch.org/whl/cu118

 

[KIWIvolshebnik]

22 minutes ago, sleighzy said:

Can you run the below commands, run each of them separately, agree to any messages which ask you if you wish to remove packages. It's fine if it doesn't locate a package, this is expected, but let me know which ones it doesn't find. The last command will then install a different version of torch (it's possible in one of the earlier versions of WhisperAttack that it installed an incorrect version. Then try both the executable and the Python version to see if this makes a difference.

pip uninstall torch

pip uninstall whisper

pip uninstall openai-whisper

pip install torch --index-url https://download.pytorch.org/whl/cu118

 

Thanks for the reply.

I uninstalled Python. Should I reinstall just to run the commands?

[sleighzy]

13 minutes ago, KIWIvolshebnik said:

Thanks for the reply.

I uninstalled Python. Should I reinstall just to run the commands?

Yes please, primarily because I don't know what Python completely cleans up when uninstalled. You may still need it if needing to use and troubleshoot with the Python version as well. You can uninstall it again once we sort out the issues and the executable version is working perfectly for you.

Feel free to jump into our Discord server as well for a bit of back'n'forwards if you like. We can post an update as to the resolution here.

https://discord.com/channels/610534461456777257/809527129422430218

Edited February 12 by sleighzy

[NAM]

Make sure that the python is version 3.10.

It took me couple hours to to get it going because mine was python 3.8.

Then I installed the latest 3.12.

Not until, I realised that whisper works python 3.10.

Everything works great much better than the default speech recognistion.

 

Thanks Sleighzy and Nikoel

[sleighzy]

3 hours ago, NAM said:

Make sure that the python is version 3.10.

It took me couple hours to to get it going because mine was python 3.8.

Then I installed the latest 3.12.

Not until, I realised that whisper works python 3.10.

Everything works great much better than the default speech recognistion.

 

Thanks Sleighzy and Nikoel

Yeah, Python 3.11 is what we have documented on the README due to that issue (its for PyTorch).

[nikoel]

Hahaha the infamous 10min, 2hr install due to instruction skimming. Happens to the best of us

[NAM]

Yep.

Skipping the README cost me couple hours of sleep and fistful of hair.

Edited February 13 by NAM

[sleighzy]

For folk wanting to use the executable version of Whisper Attack you may encounter an error stating that the cublas64_12.dll library couldn’t be found. This should be using CUDA 11 (and comes with it) but appears it’s looking for CUDA 12. If you do hit this then install the CUDA 12 Toolkit https://developer.nvidia.com/cuda-downloads

[Ourorborus]

Excellent work here.

A little concerned about the use of VRAM/Cuda reducing performance in VR. However I am in the fortunate position of a recent GPU upgrade and have a 3070 on the bench waiting to be sold. Is there a way to force Whisper onto a second GPU leaving the main one alone for graphics processing?

[sleighzy]

3 hours ago, Ourorborus said:

Excellent work here.

A little concerned about the use of VRAM/Cuda reducing performance in VR. However I am in the fortunate position of a recent GPU upgrade and have a 3070 on the bench waiting to be sold. Is there a way to force Whisper onto a second GPU leaving the main one alone for graphics processing?

Using a second GPU for this is not something we've looked at. Our default model is small.en however you can update your settings.cfg and change this to use smaller models which will use less VRAM, e.g. tiny.en

whisper_model=tiny.en

 

[sleighzy]

We are looking to release a 1.0.0 version in the near future. There is a WhisperAttack v1.0.0-alpha.3.zip (or later as we release more alpha versions) file located here: https://drive.google.com/drive/folders/1z-a-dgxeQS6Aj_jNq2mhuZNBfFwbaoNP?usp=sharing for folk to start using this now before final release.

You can download and unpack that to any location, e.g. C:\Program Files\WhisperAttack, and run the WhisperAttack.exe file (can create a shortcut to that).

Notable features in this release:

• A new window is now displayed that shows startup information, the text that was transcribed from your voice, and the text sent to VoiceAttack (or DCS only if the sentence starts with "Note"). This window can be closed and WhisperAttack will keep running. The window can be opened again from the menu in the system tray icon.
• WhisperAttack runs in the background, with an icon in the system tray to open the WhisperAttack window or exit.
• Multiple word replacements can be put on a single line in the word_mappings.txt file, separated by semi-colons. For example, gulf;gold=Golf
• Python and ffmpeg do not need to be installed

Note: This requires CUDA 12. WhisperAttack ships with this runtime and libraries. If you however see an error stating that the cublas64_12.dll library couldn’t be found you will need to install the CUDA 12 Toolkit. Only the CUDA > Runtime > Libraries need to be selected from the installer if presented with a list of components in the installer: CUDA Toolkit 12.8 Downloads | NVIDIA Developer Please let us know if you see this and we'll have a look into it.

The README.md (nikoelt/WhisperAttack at build-whisper-server-executable) contains more information (I need to convert this to a proper PDF instruction document for the final release so apologies in advance)

Please try this out, let us know if you encounter any issues or suggested features, and look forward to the upcoming final v1.0.0 release.

A VAICOM PRO folder is also on the Google drive and contains a VoiceAttack profile for Vaicom users. Vaicom Pro 3.0.0 was released recently and their exported key words are now in the same format we use so you can export those from your Vaicom database and update this VoiceAttack profile with those.

Dark mode in the pipeline for a later release for those that enjoy that.

Edited February 28 by sleighzy

[sleighzy]

WhisperAttack v1.0.0 - Major Release!
We’re excited to bring you the first WhisperAttack v1.0.0 major release with improvements, enhanced accuracy, and a brand new UI!

What’s New?

WhisperAttack runs as a Windows application

• WhisperAttack is now shipped as a Windows application.
• This runs in the background with a system tray icon that can be used to open the WhisperAttack window or exit the application.

New UI to display startup information, transcribed text, and text sent to VoiceAttack or the DCS Kneeboard

• An application window is displayed showing you all the helpful information without needing to look at log files.

Improved Installation & Dependencies

• Installation is as simple as unpacking it and running as a normal application.
• No more need for installing Python and libraries and viewing things in the console.
• Multiple word replacements can be put on a single line in the word_mappings.txt file, separated by semi-colons. For example, gulf;gold=Golf
• WhisperAttack uses faster-whisper for transcribing text. This means it is no longer necessary to install ffmpeg.
• The CUDA 12 runtime is included in the installation, this should remove the need for installing the CUDA Toolkit 12.

 

Release can be located here: Releases · nikoelt/WhisperAttack

Edited March 3 by sleighzy