CQC Voice with Acoustic Magic mic

[Shaky]

I've finally gotten around to trying this Acoustic Magic Microphone out and am running into some odd behaviors with CQC voice that I need some help with.

First a bit about my setup:
1. The Mic. plugs into an older Audigy 2 card using the Mic input plug... no USB mic. like the Kinect which I mention because it's a key difference I think.
2. Using non Creative drivers for the sound-card called KX audio.  They're infinitely tweak-able and I have used for many year as they allow 4 discrete outputs from the card for multi-room audio... They also have allowed me to really get the gain of the Mic. input to an awesome level. I don't think this is really all that relevant to the issue I see but worth mentioning.

Now, I can record my voice just fine and sounds loud and clear even from across the room.

The oddity I see is in getting CQC voice to wake up and enter dialogue mode.  Saying " Hello Jarvis"  is very poorly recognized almost like the mic. gain is turned down to 20% or something.  It usually takes 3 or more attempts speaking right up to the mic. and very clearly saying the wake up words.  Saying something like Hey Jarvis, turn on the xyz lights also just doesn't work well.

The wonderful part is that once it enters the dialogue mode and CQC voice is fully awake, it's nearly flawless with recognition even from 20+ft away. (It's like the gain is turned back to 100%).  I don't actually see the gain/volume changing though...but something odd is happening behind the scenes comparable to that.

Do you have any ideas of what this could be? 

It's also worth noting that when I use a USB soundcard and put the Mic. on that the issue of waking up CQC voice goes away... but I can't then get the gain high enough to get the good recognition at a distance.  The Kinect acts the same.  It's almost like something is different between a USB soundcard and a PCI one.

Could CQC be doing something different when in standby listening mode vs dialogue mode with a Mic. on a PCI card somehow? Something to do with exlcsuive mode/grabbing the mic. vs sharing?  (I do have allow exclusive mode checked to allow in the recording settings and col up to 100)

Let me know if you have any ideas to try.  I'm stumped.   Anecdotally  with using the Acoustic Magic mic. on a USB soundcard vs the Kinect, the Kinect seems a bit better.  What I'm finding is it's all in getting the gain of the Mic. as high as possible without clipping/distortion that seems to make all the difference.  If I could get the Acoustic Magic mic. to pick up the hello Jarvis bits as clear as it get everything else after that I think we'd have a winner.

[Dean Roddey]

Make sure that the power profile isn't allowing anything related to the audio processing go to sleep or anything like that.

I wonder if the mic has a lower power mode it goes into until it hears something? For its normal use as a conference room mic, that probably wouldn't be much of an issue, so it's possible it does that. If you just say some random sentence fairly loudly first, then do the wakeup, does it pick it up then?

It might be worth trying another sound card, as a sanity check.

[Shaky]

Make sure that the power profile isn't allowing anything related to the audio processing go to sleep or anything like that.

I wonder if the mic has a lower power mode it goes into until it hears something? For its normal use as a conference room mic, that probably wouldn't be much of an issue, so it's possible it does that. If you just say some random sentence fairly loudly first, then do the wakeup, does it pick it up then?

It might be worth trying another sound card, as a sanity check.

Yes it does have a low power mode that could cause wake up issue (LDS) switch, but I have it off.  I played around with that switch just to be sure but no changes so far.  It's deeper than just waking things up with one loud voice.  I have to say Hello Jarvis multiple times and really enunciate for it to recognise.

I have used 2 different physical sound cards passed to the same virtual machine with no change.  I think I'll try a different PC with a mic input and see if that helps.  Like I mentioned I used a USB soundcard and the wake-up issue goes away in that same VM... downside is with that USB soundcard I can't get the gain high enough to have the magical recognition I am striving for.   I'll play around some more and see what else I can figure out and post back later.

[Dean Roddey]

Does the same problem exist if you do an exchange of a couple commands, then dismiss Voice, then immediately use the wakeup phrase? If it still doesn't respond, then it couldn't be related to any sort of power down. If it does respond, then it might. So that would be a good divide and conquer sanity check.

[Shaky]

Yeah, same problem happens after you dismiss.  It resets things back to nearing impossible to wake it up again.
Did some new testing:

- On the wife's laptop I installed CQC voice and enabled the 3.5mm Microphone input, maxed out volume/gains.
- tested with a USB headset first to make sure CQC voice is working, with that is wakes up and responds as expected.
- plug in the Acoustic magic (reset it to default mic via the 3.5mm jack input.) press reload on CQC voice... says ready to help
- I can't get it to wake up at all.  
- Tried an old cheap microphone with 3.5mm jack.  Same issue, won't wake up.
- Just to make sure, I tried to record my voice with those same two mic's I tried and they are working just fine so CQCV should be hearing things so to speak.

Now I'm a bit stumped as to why I can't wake it up at all off the laptop mic input/soundcard, but have a theory.  If the cause of this is that the input volume is somehow getting drastically lowered in software for a non USB input and not re-enabled within CQC. The acoustic magic on my tweaked soundblaster may still have enough gain to be heard, although just barely which explains why it's so hard to wake up.  The ordinary soundcards however just don't have enough oomph to be heard at all... maybe?  Could somehow CQC voice be muting or nearly muting one of the mic inputs within CQC and just forgetting to re-enable or something silly like that?  I ask this because I notice the CQCV logs that the mic is enabled/disabled often during conversation mode.  That enabling may be what fully wakes it up. Is there perhaps a difference in a USB vs 3.5mm input in the windows API or whatever it is you are using to enable and disable the mic? Might explain why USB works fine and 3.5mm not.

Not sure where to go from here.  Do you have a way to test with an older analog mic on a PC with 3.5mm input vs a USB soundcard/Kinect or other onboard mic to see if you can replicate?

Let me know if anything else needs to be ruled out first.

[Dean Roddey]

CQCVoice never messes with the levels at all. That's all left to the speech recognition library, if it does anything. We create a wave input on the default audio device and just read bytes and pass them as is to the SR engine's input stream.

Try a different name and see if that makes a difference. Something maybe longer. Maybe Elizabeth or Artimus or something like that. Just for funzies.

[Shaky]

CQCVoice never messes with the levels at all. That's all left to the speech recognition library, if it does anything. We create a wave input on the default audio device and just read bytes and pass them as is to the SR engine's input stream.

Try a different name and see if that makes a difference. Something maybe longer. Maybe Elizabeth or Artimus or something like that. Just for funzies.

OK so much for that theory then although I still think there's something more than just the recognition engine going on complicating this in some way that I don't get yet. I'll try some really long names and see if that might help.  That has helped a bit in the past but for some reason I keep going back to Jarvis.