myVoice Open Beta (1 Viewer)

[booyakasha]

Yodar: There's an option in the config (TTS Debug) to have it repeat back to you what the system heard. It's great for testing. I would use only the mic that you trained it with for consistency purposes. Also, out of curiousity, how many training sessions have you had with it?

Marcusb: I think it would be great to know how it even works amongst various native english speakers, and what the nuances are. ie: British engilsh, Received Pronunciation, Estuary English, Aussie, eastern Canadian, southern US, etc. Sometimes I'm amazed this stuff works at all, as I've even had problems understanding other englishes at times...lol. I'll look into the voice profile switch. I don't think it's that difficult. I'd figure to do the same with the theme as well.

 

[Yodar]

I have had 2 training sessions.

With my headset mic its currently going good.
With the webcam with build in It isnt working.
Though I can train ok with it.

//edit it works with my web came mic. Though not verry usable.
Now im looking for a good mic so I talk from any position in my living room. Is this possible?
Ill check the mics out your described in a previuos post.

//ok, after a few training session it is now working with my webcam mic all over my living room. Only problem when sound of music is loud, it wont respond good anymore Al lot of training is realy important, dont use default profile!.


I was wondering, I think it would be great if you can say computer again right after you activated it already. So I would minamize my activation time to two seconds. And offcourse say almost computer everytime i giva a order.

Maybe this is not neccesary anymore if you have a indication the computer response is still active. As mentioned by an other user.

 

[Yodar]

I have lotsa ideas for this plugin so i hope its ok for me too put them down here. Because this is the coolest (not best) plugin I have seen so far.

One idea is too mute all computersounds when +computer is active.
This way the mic will not be troubled by audio from the speakers, beside +computer command.
Maybe this as a option which users can enable/disable.

My other idea is advanced and maybe not possible.
Compute the diffrence between of the computer audio and mic audio signals so you have:
Original audio signal
Microphone audio signal + voice.
Than substract the orginal audio signal from the Microphone audio signal so only the voice remains as input. I guess this can not be done by the plugin but in a other abstract layer in windows.
With this advanced option the computer wont be troubled by error orders from the speakers, the input should also improve while listening to music.

Spell letters,
Plugin like online videos or my wikipedia has a search field. Its much work to add a word with the remote so would be great if it can be spelled. Maybe this is allready possible to add yourself?
Im thinking of "computer" "type" A B C D etc and also type words.

//lowering instead of muting is way better

//ps, I use the eyetoy camera with build in audio. After enough trainsession I can talk all over my room and it is still working

 

[RaDiKal]

I have lotsa ideas for this plugin so i hope its ok for me too put them down here. Because this is the coolest (not best) plugin I have seen so far.

One idea is too mute all computersounds when +computer is active.
This way the mic will not be troubled by audio from the speakers, beside +computer command.
Maybe this as a option which users can enable/disable.

My other idea is advanced and maybe not possible.
Compute the diffrence between of the computer audio and mic audio signals so you have:
Original audio signal
Microphone audio signal + voice.
Than substract the orginal audio signal from the Microphone audio signal so only the voice remains as input. I guess this can not be done by the plugin but in a other abstract layer in windows.
With this advanced option the computer wont be troubled by error orders from the speakers, the input should also improve while listening to music.

Spell letters,
Plugin like online videos or my wikipedia has a search field. Its much work to add a word with the remote so would be great if it can be spelled. Maybe this is allready possible to add yourself?
Im thinking of "computer" "type" A B C D etc and also type words.

Got some great ideas here -- I'll add my own 2 cents on top...

How about lowering the volume by half after the computer hears the prompt instead of muting it? This does bring up some unintended issues such as we typically verbalize a whole series of commands to take a particular action rather than just one (move up, select item, play, etc. not to mention verbalizing the commands to raise/lower/mute the volume). Since the listen timeout is in seconds, this would leave the sound muted or low for a fixed number of seconds after giving the command. Perhaps we can have an option to go back to 'waiting for first prompt mode' through a verbal command such as "resume" so we can customize the combination to suit how we use the voice interface.

The plugin relies on Microsoft's voice recognition add-on's so the notion of doing output/input cancellation really isn't possible. I suspect MS may already be doing this to a degree within their tool. It is critically important to train the microphone exactly in the way you plan to use it. Don't train with a mic at your lips when you plan on giving voice commands to the mic from accross the room.

Calling out keyboard input as letters and numbers is a fantastic idea. Needs either an input mode command or the plug in need to recognize the virtual keyboard prompt. This is a terrific idea.

 

[booyakasha]

Some great ideas guys. I appreciate the feedback, and wish I had the time to work on it all.

I don't think the noise cancellation is do-able as I'm not an accoustical engineer, which I think I'd need to be to program that.... Good advice on the training aspect RaDiKal.

I use a 360 earpiece headset and it's quite good. I think Marcusb has a mic under his TV set, and it seems to work ok for him. Not sure on the type of mic though.

One thing I don't like about changing the volume programmatically is that I use the plugin for volume change. This would be much harder if you didn't get instant audible feedback. I'll look into doing that though, at least as a sliding scale option.

The letter idea isn't bad. However, in some of my very preliminary testing of the voice engine, it seems to have problems understanding the difference between similar sounding letters ie: "B" or "D". I had thought about using military type phrasing, such as Alpha, Bravo, Charlie, etc. Not sure if that would be too cumbersome though. Further to this though, part of my future plans are to accept dictation for this kind of input. I wanted to hammer out all the current issues first, and then re-design with that in mind (half the code is very early C# for me, and not well structured)

Yodar: You can use a wave response in your deactivation phrase to signal when the computer has stopped listening, although I know it's not as good as a visual cue (haven't figured out how to do that)

 

[Marcusb]

yep, I have a mic under my TV. It works well for the most part, except when there is audio playing from the TV.
Then it becomes "difficult". The mic is a relatively cheap desk top one. It's the best I could find without going to more pro grades.
I think I said a lot earlier in this thread that my idea for background noise cancelling is that it should be done in hardware. shouldn't be too hard with an operational amplifier and they start at about a dollar. An Operational amplifier has a postive and a negative input pin. You connect the negative to the PC audio out and the positive to the mic and the output of the opamp to the PC mic input and the PC will only see the mic inout minus the TV output. Of course there would need to be a bit of work done to balance this, as the levels have to match. Maybe some automatic gain control. It's never as easy as it firts sounds, hey? I'm just never home enough to try this out myself.

hey, for anohter one of my silly but "what if" ideas is that the response for things like date etc are hard coded. Could we have variables for this instead? you know, the usual <dddd> = day of week, <yyyy> equals year type of thing.

 

[booyakasha]

Yeah, I really like that mic idea Marcus. Too bad my electronics ability is rubbish.

That variable thing is something I think you asked me a long time ago. Never really got to it obviously...lol
I'm never sure how to implement all that though. I'd like to add things like weather to it, and generally more interactiveness.

Anyways, for anybody that's interested, attached is a version that adds a volume reduction slider to the settings screen. I think that's what you were after RaDiKal. Use a combination of that and adjusting the timeout to find the right balance. There's an issue with it however, that prevents you from adjusting your volume by voice, well having it stick anyways. I'll look into that when I get a chance. I figure I'd get it out there for you to try, as this was relatively painless to implement.

Also, you'll want to backup the dll files this one replaces.

 

[Yodar]

Offcourse, dont let us rush you with new ideas.

The mic operational amplifier is a good idea. Beside you have to create some electronics
Only since I dont have acces to an osciliscoop and function generator its a bit out of the question, for now.

Though the idea to slide sound is really important i think. When i go to My tv the computer start getting random orders pretty fast.

I also was thinking about an "one time" commands.
Commands that automaticly trigger "computer stop listening".
Example, if I want to turn on the light "computer turn light off", "computer turn light on" So i dont have to say computer stop listening so often. It would be great if you can just check an option. In the same screen as where you add the sounds, ID number etc.

 

[bruddha]

one time commands and album titles

hi all

the "one time" command idea would be a good extension for this plugin. i'm annoyed by the random commands. especially when debug mode is on i end up in an endless babble until i switch off the microphone.

another question? couldn't you add the album names in the word list, instead of only artists, titles and genres. i know your already answered that the quality of album tags are bad sometimes, but let the user decide himself.

thanks alot for this great plugin.

 

[booyakasha]

To emulate the one-time command, can't you just decrease the timeout to 1 or 2 seconds or so?

Albums, video's etc, are at the top of my todo list.