WhisperAttack - OpenAI Whisper Voice Recognition and injection into VoiceAttack

[VirusAM]

7 hours ago, sleighzy said:

Hi @VirusAM. Can you enable debug within Vaicom and then replicate this issue and then drop your C:\Users\username\AppData\Local\WhisperAttack\WhisperAttack.log file here. Please also provide a screenshot from VoiceAttack so I can see the full output from the debug. Can you also ensure that the Vaicom window is closed, i.e. completely closed so that it is just in the system tray.

Can you also try this by updating your settings.cfg file in WhisperAttack so that the whisper model used is tiny.en (I think it's small.en by default).

I see in your screenshot the message "PTT: use an active TX node" when this occurs. Vaicom has a timeout from when you release the tx button till it receives the actual command. This means that VoiceAttack gets the command, sends it to Vaicom, but due to it taking too long it won't do anything in the game. It seems like maybe there is a timing issue and it is taking too long from when WhisperAttack hears your command until VoiceAttack processes it after you release the button. What video card do you have? I have only seen this with one user before, it seemed to be a VoiceAttack issue of some description at the time. Are you on VoiceAttack v1 or v2? There is a profile tweak we can make that may help but would be good to see the whisper attack log file for this issue first to confirm.

You also mentioned that the default Windows speech engine works just as well for you as WhisperAttack. If that engine is working for you then WhisperAttack may not be needed, this was just beneficial for users for whom the Windows one doesn't work well. There is no ability to train this whisper model.

Hi..I will do what you ask tonight after work.
I have a 4090, and I am using the default small.en model. 
Yes I have noticed the "PTT: use an active TX node" message. That happens only when I have DCS in the foreground. 
I would like to continue to use WhisperAttack...my speech engine is trained a lot, I just tried something that in Vaicom worked well also before, but there is a lot of stuff I couldn't use well because the windows speech engine didn't recognize me (Airio stuff for example)

 

[Fox-one-76]

Hi Sleighzy

I just did as you asked, 

i found an old version of the VAICOM folder in the APP folder : I deleted it

I removed all other instances of VAICOM profile so basically i have  only 1 profile and that is VAICOM.

the problem is still here.

i took the liberty to try out what you asked VirusAM....

i have a 4090 as well.

voice attack version check tells me i am up to date.

I play with VR, and i noticed that when DCS windows has focus the problem is here :

WA does not seem to process the voice (start recording, stop recording) so nothing gets passed to Voice attack

BUT

As soon as i ALT-TAB out of DCS ("windows out of focus" appears in DCS on the VR headset),  WA recognizes instantly the previously recorded sentence (raw transcribed text "request startup" , sent text to voice attack "request startup") but of course VA has already timed out...

This behaviour happens both with VA and WA minimized to systray  and WA and VA windows visible.

For what it's worh i am also running DCSBIOS / open tablet driver / openkneeboard /  pimax play / Streamdeck  from elgato

Hope this helps

 

Edited May 26 by Fox-one-76

[sleighzy]

WhisperAttack v1.2.0 is now available.

What’s New?

• WhisperAttack now has a new screen for entering word mappings. These take effect immediately so no more needing to update the word mappings text file and restarting WhisperAttack.

Fixes & Improvements

• Fixed an issue where the transcribed text could add unexpected dashes (- and --) at the end of words and sentences, these are now removed.
• Minor UI improvements

Release v1.2.0 · nikoelt/WhisperAttack

null

On 5/3/2025 at 2:45 PM, FredOfMBOX said:

Thanks for this. WhisperAttack is a lot less frustrating that the default voice recognition.

I have an issue, though. I keep getting output that ends in "--":

Starting recording...
Stopped recording
Raw transcribed text: ' F1-- (static)'
Sent text to VoiceAttack: F1 static
Starting recording...
Stopped recording
Raw transcribed text: ' F1'
Sent text to VoiceAttack: F1
Starting recording...
Stopped recording
Raw transcribed text: ' Communication menu'
Sent text to VoiceAttack: Communication menu
Starting recording...
Stopped recording
Raw transcribed text: ' Communications menu--'
Sent text to VoiceAttack: Communications menu--
Starting recording...
Stopped recording
Raw transcribed text: ' Communications menu--'
Sent text to VoiceAttack: Communications menu--
Starting recording...
Stopped recording
Raw transcribed text: ' Communications menu--'
Sent text to VoiceAttack: Communications menu--
Starting recording...
Stopped recording
Raw transcribed text: ' Communications menu--'
Sent text to VoiceAttack: Communications menu--

It's not every time, but it's frequent. I could go into my VA profile and add `--` as an optional alternate, but I'd rather avoid that if possible. Any thoughts?

@FredOfMBOX , this is now fixed in the latest v1.2.0 release.

[Fox-one-76]

Hi Sleighzy

first of all thanks for the new release...

i just installed it  and did some tests..

sadly the problem is still there....

i did some tests using 

MB339 and C101

and also the A-10CII for a less "exotic" aircraft..

all exhibit the same behavior...

when the DCS window has the focus.... nothing gets transmitted between WA and VA

When DCS windows does not have focus everything works as advertised....

any ideas...?

thanks ...

[sleighzy]

6 hours ago, Fox-one-76 said:

Hi Sleighzy

first of all thanks for the new release...

i just installed it  and did some tests..

sadly the problem is still there....

i did some tests using 

MB339 and C101

and also the A-10CII for a less "exotic" aircraft..

all exhibit the same behavior...

when the DCS window has the focus.... nothing gets transmitted between WA and VA

When DCS windows does not have focus everything works as advertised....

any ideas...?

thanks ...

Hi @Fox-one-76. I've tried numerous things but am still unable to reproduce this. Can you confirm that you aren't running DCS as administrator, e.g. right-click both your shortcut for DCS, as well the DCS.exe file being used, and then select properties and make sure "Run as administrator" isn't ticked. I also run with OpenKneeboard and Open tablet driver so wouldn't expect an issue there. Can you try without the streamdeck and DCSBIOS? This issue occurs in both VR and 2D correct? I'll re-run all my tests again in VR but would expect this issue to exist across both.

WhisperAttack transcribes the speech-to-text and then executes VoiceAttack with this, which will reuse the existing VoiceAttack if its already running. It's almost like this part of it is being blocked until focus is taken away from DCS. For example, when you start up VoiceAttack for the first time do you get a message "Do you want to allow this app to make changes to your device?".

Is this VoiceAttack v1 or v2?

I'm assuming this profile is configured with "Send this command to active window".  (which is correct)

null

[Fox-one-76]

Hi Sleighzy

thanks again for your time

I was using voice attack V1.16

i upgraded to voice attack V2 and the issue is now gone...

it seems it was all related to voice attack version 1

many thanks for the help and your work...

[VirusAM]

8 hours ago, sleighzy said:

Hi @Fox-one-76. I've tried numerous things but am still unable to reproduce this. Can you confirm that you aren't running DCS as administrator, e.g. right-click both your shortcut for DCS, as well the DCS.exe file being used, and then select properties and make sure "Run as administrator" isn't ticked. I also run with OpenKneeboard and Open tablet driver so wouldn't expect an issue there. Can you try without the streamdeck and DCSBIOS? This issue occurs in both VR and 2D correct? I'll re-run all my tests again in VR but would expect this issue to exist across both.

WhisperAttack transcribes the speech-to-text and then executes VoiceAttack with this, which will reuse the existing VoiceAttack if its already running. It's almost like this part of it is being blocked until focus is taken away from DCS. For example, when you start up VoiceAttack for the first time do you get a message "Do you want to allow this app to make changes to your device?".

Is this VoiceAttack v1 or v2?

I'm assuming this profile is configured with "Send this command to active window".  (which is correct)

null

The issue seems similar to what I reported a couple of days ago (I still didn't have time to try again).
Maybe buying voice attack v2 is the reply for me as well?

[sleighzy]

11 hours ago, Fox-one-76 said:

Hi Sleighzy

thanks again for your time

I was using voice attack V1.16

i upgraded to voice attack V2 and the issue is now gone...

it seems it was all related to voice attack version 1

many thanks for the help and your work...

 Weird. I'm on v1.16 as well as others. Might have just been something with that particular install (permissions? ). Oh well, gremlins. Happy flying then! Sing out if you have any other issues.

[sleighzy]

9 hours ago, VirusAM said:

The issue seems similar to what I reported a couple of days ago (I still didn't have time to try again).
Maybe buying voice attack v2 is the reply for me as well?

Possibly... Could be a VoiceAttack install issue as well. Could always try v2 as well then (it's $10, but it's your money).

Do you use OpenKneeboard? You could add VoiceAttack as a Capture Window to that and then ingame see if you get anything logged, e.g. the active tx node message to see if a delay issue. If identical to Fox-one-76's issue then may not be able to be something I can replicate or fix on my end. If it's just a delay, and not a window focus issue, then I can provide a profile workaround to test.