VoiceAttack

Issues => Issues => Topic started by: bashpr0mpt on February 28, 2018, 09:07:31 pm

Title: Atrocious recognition and feedback issue.
Post by: bashpr0mpt on February 28, 2018, 09:07:31 pm
I've been using VA for a few years now, since the first voice packs came out for Elite Dangerous and I use a HCS voice pack (ASTRA ) with it. I've used this system extensively for speech recognition in the past, and it is well trained. I have run as many training sessions as my boredom level can suffer every time I open VA in hopes it will be less terrible. However, there lies the problem.

Windows speech recognition programs interface with the speech recognition system. If it makes a mistake, I correct that mistake, and it will be unlikely to happen again. But when I say;

"Hello Astra."

And it hears;

"Stop engines."

Then I am pretty much fekked. I've excessively chatted away to this thing only to see the most absurd word guesses in the history of speech recognition turn up. It's literally like I've left my audio channel open and am playing ten YouTube videos in the background at once. Random ass words that would never be used together.

VA works good for VERY. SIMPLE. commands. I often find myself having to repeat commands. But overall because it cannot learn from it's mistakes it makes absolutely ridiculous mistakes so damn often. Now with commands that include moving your ship being entered into it I can see thousands of users around the world getting absolutely destroyed by this program.

Is there any way to force it to stop being psychotic in word selection? Here is an example of the abortion of recognition it perceived whilst I was reading a Windows Speech Recognition training session. It's accuracy is about 50%, if you remove words like 'if' and 'the' etc its accuracy would be significantly lower. What in the settings can I possibly change to fix this?

Edit: Just to clarify I am using a Rode NT-USB studio microphone, and have no ambient background noise (I even use a silent keyboard and mouse as I record stuff from this machine and don't want background noise.)
Title: Re: Atrocious recognition and feedback issue.
Post by: Gary on February 28, 2018, 09:22:01 pm
Hi, there.  Sorry to hear you're having so much trouble with VA.  Windows' speech recognition, although not great for dictation, is really really great at recognizing short stuff (like you've said).   That is, unless your system isn't up to par audio-wise.  You'll want to take a long walk through this thread, as there's something up on your system that is impeding performance: 

http://voiceattack.com/SMF/index.php?topic=1635.0

That thread is a collection of pretty much all of the audio help over the years, and will probably hold your solution.  I'd take a look at trying a new speech profile, since you've indicated you are using a table mic.  I'd also look at the command weight and make sure it's not really high.  I keep mine at 50 (default). 

Hope some of that helps!



Title: Re: Atrocious recognition and feedback issue.
Post by: bashpr0mpt on February 28, 2018, 11:26:11 pm
http://voiceattack.com/SMF/index.php?topic=1635.0

I was about to add after my first edit that I went through http://voiceattack.com/SMF/index.php?topic=1635.0 before writing, and it totally slipped my mind. One thing I've noticed command weight and minimum confidence levels tool tips are a bit ambiguous, I suspect they may be a key to getting it to work better; I have them set to the default. Should I change those?

Also is there any way to correct mistakes so the system can learn and adapt with non-critical command errors where we have the time to do so?

At present the only REAL thing I can think of that would improve accuracy is to turn on speech recognition, open wordpad, and boringly read through the commands over and over again that I want it to get better at - but surely there's got to be a better way than that? Also why hasn't anyone anywhere in the forums even suggested that method? Wordpad is linked to Windows onboard SR, so correcting errors there WILL 'teach' your speech profile.
Title: Re: Atrocious recognition and feedback issue.
Post by: iceblast on March 01, 2018, 05:39:21 am
Have you tried installing Microsoft Speech Platform 11.0. You don't have to train it, and seems to understand most things said with very little issue. The default windows speech recognition never worked very good for me, but Microsoft Speech Platform 11.0 worked almost perfectly from the moment I installed it.

I don't think you can use VoiceAttack Dictate command with 11.0 though, but considering, you can't use it as it is....
Title: Re: Atrocious recognition and feedback issue.
Post by: ajg007 on March 03, 2018, 10:25:28 am
What does voice attack do when you say the words.    Hello crewman astra

My issue was not using the correct commands.....
Title: Re: Atrocious recognition and feedback issue.
Post by: Assimilator1 on March 04, 2018, 08:05:11 am
I don't know if this is the same issue I had (Gary - yes I finally fixed it! :D I'll post more in another thread), but 1 thing I found out through my investigations, is that when training the windows speech recognition (for Win 7) it sets it's own mic volume. Which for some reason was at 100% for my new USB headset!

But with my new headset that was far too high, it was picking up my TV across the room, sometimes even recognising some words! Lol. Anyway, I set the mic vol down to 40-50%, but I have to do that every time I run VA as it seems to pick the vol that the speech engine wants!

Anyway, this means (without doing the above) the mic vol could be far too high or low & messing up your voice quality. Also, I don't suppose your mic has any kind of vol auto gain in quiet times does it?
If you consistently talk from the moment VA is started, does it accurately recognise your words then?
Title: Re: Atrocious recognition and feedback issue.
Post by: Pfeil on March 04, 2018, 02:13:28 pm
I set the mic vol down to 40-50%, but I have to do that every time I run VA as it seems to pick the vol that the speech engine wants!
The speech engine will set your microphone volume to what it was during training.
If you can't lower the volume during training, you could use a command that runs on profile load that runs a "Set Audio Level" action to automatically lower the volume.
Title: Re: Atrocious recognition and feedback issue.
Post by: iceblast on March 04, 2018, 04:00:50 pm
The other thing you can do is, when your setting up the mic in windows, it will have you speak some words, as you do that, it will adjust the volume on the mic. When it gets to what you think it should be 40-50% finish the program without saying anything. So the mic doesn't go up anymore. After that, your problem should be solved.

The same thing happened to me, so I ran the setup for the mic, and did whatever I could to fool it to set the mic volume at what really worked for me, and not what it thought worked. After that, I was good to go.
Title: Re: Atrocious recognition and feedback issue.
Post by: Assimilator1 on March 21, 2018, 11:51:11 am
Thanks for the replies guys :)
Title: Re: Atrocious recognition and feedback issue.
Post by: iceblast on March 21, 2018, 01:53:42 pm
I found a program, that will allow you to set your mic to whatever lvl you want, and it won't let anything change it.

I set it for 50% Mic Vol LVL for you, if you need to change it read the How to.txt