VoxCommando
Microphones and Speech Recognition in General => What is a good microphone for speech recognition? => Topic started by: philby85 on January 17, 2014, 02:06:44 AM
-
Ok, I looked on ebay for second hand 360 kinect. Cheapest I could get was 69.95 plus 9.95 postage (AUD) then I had to buy a power adapter with USB cable which was 20.00 delivered. All up 100.00.
I checked with my local game store and I picked up one with cables and adaptor for 42.00 Bargain !
I downloaded SDK 1.8 installed and tested. Seemed to be hit and miss and I had to stand right in line with it, but for what I paid no problems.
I read through the forum and found a suggestion to train it a few times. Wow ! I can be anywhere in the room and it will pick me up. I am getting between 90-96 % accuracy.
My boundary mic was good, but I was never able to get it to recognise me while media was playing, the kinect can, most of the time.
So, impressed so far.
cheers
Phil
-
Wow seriously? I tried this way way back so maybe old version of drivers/sdk
But it was absolutely atrocious.
Bad accuracy, bad input, tons of ambient noise and forget if something was playing.
May have to give this another go sometime
-
I also had pretty disappointing results with the Kinect. It could be the drivers, or it may come down to quality control and a difference in the actual microphone array on each unit.
-
Hi everyone. This is my first post on the forums, and to be honest the first time I have posted in a forum for some time. I saw this post and wanted to give my experience as it's been quite positive.
I have a 360 version of the kinect sensor that I am using alongside vox commando and it has the best reception and clarity for vox commando of any device I have used with the software. I am not 100% sure if this is because of newer drivers etc but it's been great using the kinect for this purpose.
My understanding is that it uses an array of microphones giving it much greater redundancy than a dedicated microphone. I had decent success with my headset mic which is a microsoft lifechat lx 3000, but the kinect blows it out of the water, in fact I was stunned (as a techie) on how far away from and how low I could speak and still have commands understood without having to repeat them etc.
I was stood outside the room of operation at the top of the stairs which is a good 20ft or so and beyond the doorframe to the bedroom. I had no idea anyone had any problems with them, just saying my experience with the kinect 360 sensor has been stellar.
In regards to the ambient noise and not hearing people over whatever is playing, I am using a program called FAAST, and once I have it sorted the plan is to simply hold my hand forward to ask it to listen for commands via vox commando by coming out of stand by and also reducing the volume or pausing the movie.
-
Here as well I am having positive experience with Kinect ... and compared to array mics it is a pretty good deal even at 100 (got mine on ebay for 60 +10 for power supply, all delivered)
In regards to the ambient noise and not hearing people over whatever is playing, I am using a program called FAAST, and once I have it sorted the plan is to simply hold my hand forward to ask it to listen for commands via vox commando by coming out of stand by and also reducing the volume or pausing the movie.
I have been looking for a similar solution ... can you elaborate about FAAST installation ?
thanks
-
Well at the moment I am not having much in the way of success with any other method of visual gesture control in xbmc, kinemote is hard to find a download for. I am using kinemote light and well lol, that's just pathetic from my standpoint.
I have also tried a few others, kinect remote (kinemote's ancestor) and the kinect adapter for xbmc. The last of which had the most if limited success.
FAAST, has a very nice interface for seeing what the kinect is seeing (skeletal data, depth data etc) and you can switch between upper and full body modes. It works by setting up a series of events that trigger key presses and is by far the most successful program I have used so far for the purpose.
I have used it in one game so far (crysis 2) in which I used another voice app + FAAST to change weapons, shoot, jump etc. I will actually upload a video to youtube to demonstrate if that's allowed on the forum.
If you want to take a closer look go to http://projects.ict.usc.edu/mxr/faast/ (http://projects.ict.usc.edu/mxr/faast/)
Or take a look here https://www.youtube.com/watch?v=EZNBw_0fVFk (https://www.youtube.com/watch?v=EZNBw_0fVFk)
I think so far this will be the best option (for me) to get a workable visual control system working without issues getting in the way. That combined with vox commando should be a powerful combination. Will post a video on my progress as soon as I can.
-
Those of you with Kinects -- have any of you installed the latest version of VC2 (2.1.0.3) and tried the new experimental Kinect audio settings?
http://voxcommando.com/mediawiki/index.php?title=ChangeLog#Version_2.1.0.3
-
I tried the new kinect feature a bit ...
I did not have the beam angle event triggered ...
I can't say that I noticed a difference. Specially since I added sound filters (noise gates, eq and AGC) I get almost always more than 90% confidence true positives ... And those that are not get handled by the adaptive confidences functions
As far as I know if I choose the kinect stream it will by pass the sound filters .... If there is a way I will do more testing
-
Well I guess it is not of any use to you personally.
Beam events are not activated yet, but I'm not sure how useful they are anyway to be honest.
Acoustic echo cancellation, noise suppression and beam forming are all built in if you enable audio streaming, so no, you cannot and in any case probably should not apply additional third party manipulations to the audio.
-
Automatic gain control is also an option but it is generally not recommended for speech recognition.
If you do use Kinect for speech recognition you should probably set the input level to 10 percent or less
-
Great ... Will give it another try .....
Question: can I run 2 instances Vox commando (from different folders) one is Spanish using kinect and one is standard in English ? Or both have to use SP ?
-
Beam angle can be very useful in open space implementation
As if you say 'turn on the light' beam angle can tell VC that the command came from the dinning area... Or kitchen or sitting ... Etc . For sure that will depends a lot on the room arrangement
-
Great ... Will give it another try .....
Question: can I run 2 instances Vox commando (from different folders) one is Spanish using kinect and one is standard in English ? Or both have to use SP ?
Yes, to run two or more instances of VC, create for both a link on the desktop. Go to properties and put at the end of the target -new
You can use for both different settings - in this way I use english for music/video and german for home automation.
see example picture
-
Beam angle can be very useful in open space implementation
As if you say 'turn on the light' beam angle can tell VC that the command came from the dinning area... Or kitchen or sitting ... Etc . For sure that will depends a lot on the room arrangement
Sounds good in theory. I found the beam angle was constantly changing in my initial tests. Rather annoying.
-
My tests with the developer toolkit was pretty good ... Might you kinect be with some issues ?
-
Maybe you are have a different Kinect model ::hmm
"Kinect Models
As of 12/14/2013, there exist four different models of the Kinect:
The original Kinect-for-Xbox, released in November 2010, with model number 1414. Until version 2.8 of the Kinect package, this was the only supported Kinect model.
The new Kinect-for-Xbox, model number 1473. This was an upgrade to make the Kinect-for-Xbox compatible with the then-new Kinect-for-Windows. First supported in Kinect package version 2.8.
The Kinect-for-Windows, model number 1517. First supported in Kinect package version 2.8.
The Kinect 2.0, released together with the Xbox One in November 2013. Not yet supported by the Kinect package.
The main difference between models 1414 on the one hand, and models 1473 and 1517 on the other hand, is the arrangement of the Kinect's sub-devices (camera, microphone array, tilt motor, internal USB hub). In model 1414, the camera was the "main" device, and had the Kinect's serial number attached to it. In models 1473 and 1517, the serial number is now attached to the microphone device, and the camera device's serial number is bogus."
copied from this link: http://idav.ucdavis.edu/~okreylos/ResDev/Kinect/MainPage.html (http://idav.ucdavis.edu/~okreylos/ResDev/Kinect/MainPage.html)
-
I have a 1414. I bought it new.
-
I have a white one model 1473
-
To all of you using a kinect how do you get around the sound from music or a movie interfering? For example if I am watching a movie via xbmc the OSD keeps popping up. Of course no commands are triggered because I have prefix mode on. Still annoying though to have it pop up every few seconds.
-
Uncheck "Show OSD on Command" in Options.
http://voxcommando.com/mediawiki/index.php?title=Options
-
If you have version two you might want to use this: http://voxcommando.com/mediawiki/index.php?title=Options#Only_show_OSD.2FAlternates_if_command_is_executed
-
Thanks James and Naomi. Much appreciated.
-
Yes, to run two or more instances of VC, create for both a link on the desktop. Go to properties and put at the end of the target -new
You can use for both different settings - in this way I use english for music/video and german for home automation.
see example picture
Thanks kalle ... Just saw you post today ... Will give it a try on the weekend
-
Thanks kalle ... Just saw you post today ... Will give it a try on the weekend
The command line documentation on our wiki needs a bit more fleshing out, but you may still be interested in this link: http://voxcommando.com/mediawiki/index.php?title=Command_line
-
I have done more testing with kinect audio stream ... It is really good most of the time I get 90% confidence ...very comparable to my normal setup.
However, In my case, I need noise gate since I live close to the street and railway ...
I do not know if kinect has such functionality ... In normal setup, I have a noise gate with different bands, then equilizer, then AGC applied to kinect mic then passed to VC
I hope I will be able to use the beam angle event without using audio stream
-
However, In my case, I need noise gate since I live close to the street and railway ...
I do not know if kinect has such functionality ...
What you see is what you get. You should enable AEC (maybe) and disable AGC (almost certainly). If that is not good enough then you need to stick with your existing setup.
If you use AGC with Kinect it will make it more likely to pick up background sounds when you are not speaking.
Also you should verify that your Kinect input level is less than 10%.
Either way you will of course still need to use prefix/standby mode.
I hope I will be able to use the beam angle event without using audio stream
You won't. Not unless you create a separate program to track it and send events.
-
Haddood, can you please try the Kinect again with the alpha I just uploaded:
Version 2.104. It's on the download page.
There was a bug that may have prevented proper kinect audio streaming. I've also enabled the beam angle events.
If it doesn't work well please post a log.
Note that I think you must do a full restart after enabling beam events in options.
-
But instead of beam angle events, I think I might include an action that returns the last known beam angle.
-
Haddood, can you please try the Kinect again with the alpha I just uploaded:
Version 2.104. It's on the download page.
There was a bug that may have prevented proper kinect audio streaming. I've also enabled the beam angle events.
If it doesn't work well please post a log.
Note that I think you must do a full restart after enabling beam events in options.
I will do tonight and post back ....
But instead of beam angle events, I think I might include an action that returns the last known beam angle.
that is even better :)
by the way, just got my second kinect 8) 8) model 1473 got really a good deal on ebay ... so I should be able to test both models (2nd to go on second PC, that will take few days preparation)
-
I installed 2.104 ... works like a charm and disabling AGC makes it even better (less false positives)...
event beam angle are generated ..... however in their current setup I don't know how to use them, due to the following issues:
- they trigger regardless if there is a recognized command or not ...
they trigger after the recognized command ... so it is to late to get info from them ..
I am not sure if it is doable but if they trigger before the command we can set a command that assign the angle to a variable or as you mentioned make the angle accessible via variables to be used with the actions of the recognized command
and that would leave room for further actions based on the angle. like shift the speakers balance to your side before speaking... that would make VC amazing as it will feel it is talking to you ::) ::) (or just mount dedicated TTS speaker on a rotating base and have it turn towards you 8) 8)... )
anyway ... it is working great ... will keep using it for the time being and see if any issues comes ... if there is any specific testing you need let me know
-
event beam angle are generated ..... however in their current setup I don't know how to use them, due to the following issues:
- they trigger regardless if there is a recognized command or not ...
they trigger after the recognized command ... so it is to late to get info from them ..
I am not sure if it is doable but if they trigger before the command we can set a command that assign the angle to a variable or as you mentioned make the angle accessible via variables to be used with the actions of the recognized command
The beam angle event has no direct relationship with recognised commands.
The beam angle event is triggered when the beam angle change is detected. It is not possible to change when that happens. I am surprised that it does not fire until after you finish giving your voice command, I will have to do some tests myself.
I will also add an action to return the last known beam angle but based on what you describe that value may not be up to date enough.
-
event beam angle are generated ..... however in their current setup I don't know how to use them, due to the following issues:
- they trigger regardless if there is a recognized command or not ...
they trigger after the recognized command ... so it is to late to get info from them ..
I'm not sure how thoroughly you tested this before posting your results, but I tested this again a few times and the beam event always fires before I finish the first word of my voice command. Certainly long before my commands is recognized.
In any case I've uploaded 2.105 (2.104 had a bug in TTS.SpeakSync) and this version now has a couple of Kinect actions in it including one to get the current beam angle. This action should work even if you don't enable the beam events in options.
-
I'm not sure how thoroughly you tested this before posting your results, but I tested this again a few times and the beam event always fires before I finish the first word of my voice command. Certainly long before my commands is recognized.
I have done more tests. created new copy of VC folder ..emptied the commands except for one ... it turn out you are right, the beam event does fire before the vc.reco. however sometimes it does not fire at all, then fires for whatever noise is around after the command which led me to conclude it is after ... see the attached log. line 620 and 700 (if opened in NotePad++) the commands do not have kinectbeam event before.
after this post I am conducting another test, I will put back my full configuration file again. and see if the same things happens and I will post here..
just for the remote possibility of influence, I am running VC on a MAC mini 2006 core2 T5600 1.83GHz with 2GB ram and an SSD. when my full configuration is loaded VC will take about 100MB of RAM, while in the test setup above (1 command) it is 50MB ...
-
in this post I am attaching the log of my full configuration ... it is a bit a jungle in there :-\ :-\
I issued the same command as in the first test (bixi info) ... twice the first time it fired Kinect event before ...the second did not
you can find the entries in line 3899 & 4084
will switch to version 2.105 and more testing tomorrow
-
Hi Haddood, I have one question.
Did you tried your test always from the same position?
I noticed that the event is only fired if the beam angle has changed and can confirm James statement that the event is fired before VC recognized the command. So if I speak my command when I sit in the front of the kinect twice, the beam angle event will show in the history window only one times, because both command come from the same angle.
I've tested this in VC 2.104 from a short distance. Now I will do a test with VC 2.105 :)
-
it turn out you are right, the beam event does fire before the vc.reco. however sometimes it does not fire at all, then fires for whatever noise is around after the command which led me to conclude it is after ... see the attached log. line 620 and 700 (if opened in NotePad++) the commands do not have kinectbeam event before.
It looks like it is working exactly as intended and expected.
-
Hi Haddood, I have one question.
Did you tried your test always from the same position?
I noticed that the event is only fired if the beam angle has changed and can confirm James statement that the event is fired before VC recognized the command. So if I speak my command when I sit in the front of the kinect twice, the beam angle event will show in the history window only one times, because both command come from the same angle.
I've tested this in VC 2.104 from a short distance. Now I will do a test with VC 2.105 :)
I was moving all the time. Maybe that's why when I Issued, is from the same place I thought and every case that event did not trigger every time .. In addition I wasn't aware of James intention of firing the event only when the angle change...
-
I have been using the kinect feature for a while now ... And it is working really good ... With AEC it is doing great
So far the best results for open mic
-
cool, glad to hear it!
-
HI all;
I am new to Voxcommando and I'm really anxious to play with my 360 kinect with it. I have been racking my brain for 4 days trying to make it work. I am hoping someone can point me in the right direction to fix the problem.
I have Voxcommando running on Windows 8.1. I have a AMD 8350 processor with the asus a578-usb mother board.
I've installed the 1.8 sdk and the developer tools 1.8, in my device manager it shows the windows kinect devices with no problems everything installed correctly. Under sound card it shows a usb kinect device but when I goto to sounds under control panel it does not show the kinect microphone installed.
I can bring the camera up under the developer tools but I just cannot make the mic work. Does anyone have any suggestions as to what I maybe doing wrong or a piece of software that I need to install to make it work?
Any help much appreciated!
RJ
-
Hi RJ,
It sounds like you did everything correctly. It is really not complicated to set up. i don't even think you need to install the developer toolkit. Maybe there is something wrong with your Kinect device.
Do you have the correct power adapter to supplement the USB power? You can't get enough power from a regular USB port alone for the Kinect 360. That's the only thing I can think of that you might have done wrong.
You could try to uninstall and try again. Install the sdk before you connect the Kinect to your computer.
Maybe someone else will have some ideas. Good luck.
-
right click on the speaker in the tray, choose recording devices... do you see the Kinect mic there?
if not try right click on the white area and make sure that both show disabled, and hidden are checked ...
still you do not see?
-
Hi Haddood
Just wondering if you were able to test out Kinect versions 1414 and 1473 and if so, did you notice a difference between them.
Thanks
-
@ wacind, unfortunately not... I completely forgot about that ... all my testing was done 1414
now it will not be that easy as I changed all my setup to work with a mixer and boundary mics
-
Why did you stop the Kinect? I am just starting trying to use it.
-
Your question is not 100% clear to me. Whom are you asking and what do you mean by "stop the Kinect"?
-
Why did you stop the Kinect? I am just starting trying to use it.
If you are asking me... The reason I switched, is because I wanted to have a mic in each room in my apartment ... and kinect is great in a one room scenario
-
Hello lovely community.
Thanks to Kalle :hugs I've got my own Kinect to play around. Its a amazing piece of senorphalancs. ;)
Now I have a "liitle" question.
In the VC main window is that little nice level indicator. Is it possible to read the value from this indicator may be permanently? I want to try out, to look for the last 5 seconds or so (this is one challenge) to the level, and if its raise or fall over/under a threshhold to adjust the mic inputlevel.
On other hand I can imagine, that when a very high level is reached -sometimes I hear loud music ;) -, to switch to gesture input via kinect and trigger that way the VC-XBMC ducking procedure. This also could be nice to pause an very loud Äktschnmovie.
with greetings, fishware
-
You will need to find a third party solution to read the input level. I recommend you set it to a low level and forget it for now. There is a setting in options on the kinect tab for AGC (automatic gain control) which adjusts the input level automatically but it is generally recommended to not use this with speech recognition.
-
@Hadood.
Can you please tell me about your current setup with a mic in each room. I can barely get a mic to sound right unless it's a headset and have been doing alot of research.
How is your setup going now. Do you still get home accuracy. Omni mics are really a hit and miss. I bought 3 brands and they all have trouble picking up my voice let alone get windows speech recognition to even start voice tutorial and mic calibration
-
Hi deco,
I have seen long time ago a guy which has used a sure mic automixer for a multiroom voice control.
I can't find the website at the moment, but I know the 8 channel mixer (which is expensive, maybe there exist a 4 channel version or you can find one on ebay)
http://www.markertek.com/product/scm810/shure-scm810-eight-channel-microphone-mixer-automixer?utm_medium=shoppingengine&utm_source=googlebase&utm_source=google&utm_term=Product_Target&utm_campaign=Shopping+-+Pro-Audio&utm_medium=cpc&utm_content=kMpqojDl%7Cpcrid%7C74923786580%7Cpdv%7Ct%7C&gclid=CMPJm4zFuc8CFYZbhgodgaQIjg
-
How are all these mics that you have tried connected to your computer? Give us the exact models of the microphones you are talking about. Many microphones are designed to hear you only if you are within 1 meter or less.
Maybe the problem is the audio input on your computer.
Listen to a recording of your voice, what does it sound like. Feel free to upload a short sample.
If you are using analog microphones you might need to use a preamp or try using a USB microphone instead.
That said I personally do not bother much with open microphones because they pick up too much other sound.
-
Thank you for replying.
Think that is a pricey experiment just to have bad results.
I am currently trying different mics and just got a second hand kinect. Its exactly what I want in my living room. I hope it works as intended.
Has anyone tried setting multi kinects in each room and running two or three instances of Vox?
My last resort will be to turn to Google Home http://www.theverge.com/2016/5/18/11688376/google-home-speaker-announced-virtual-assistant-io-2016
I love their speech recognition and an api to use their mics would make vox a beast
-
I guess you didn't notice my post.
But Google Home should really be your first choice. It will be much easier to use than VoxCommando, and probably much better all around. It's also going to make it a lot easier for Google to know absolutely everything about you and your whole family, which is really important so that you can get access to the best possible ads.
You just have to be patient since you can't buy that yet. And of course if you live in some country other than the USA or don't want to use it in English, then I guess you will need to be really, really, patient.
:biglaugh
-
Sorry, never saw that post.
I dont see the fun in that at all. I have to say ok google all the time and it wont interact.
Vox is forever learning because you customize it to what you want and who you are. Google uses algorithms to "guess" what you interests are. Its not smarter if it always has to be in the cloud because once it disconnects from the internet, Its usless. I like so many have modems or other connection issues.
Take my smarthings hub. An amazing product and potential with so much problems because the company cant wait to roll out an update every few months for ''security reasons'' that has now caused my hub t disconnect and reconnect constantly due to a firmware issue. With eventghost and Vox. I can pretty much do almost anything and the only downside is the mic and lack of coding experience. Internet loss limits Vox to the WWW but not to a local network thatcan still do things around the house
James, I would take your and Vox anytime over the cloud.
-
@Hadood.
Can you please tell me about your current setup with a mic in each room. I can barely get a mic to sound right unless it's a headset and have been doing alot of research.
How is your setup going now. Do you still get home accuracy. Omni mics are really a hit and miss. I bought 3 brands and they all have trouble picking up my voice let alone get windows speech recognition to even start voice tutorial and mic calibration
my system is ok (and what is ok is a very subjective evaluation)... depends a lot on the background noise when I speak ... at night when it is very quite the system is at its best... lately working a lot, so rarely at home ... my usage dropped mainly to controlling lights, weather ...
for my setup I use
I use XAP800 ...
open mic is not easy path, it takes a lot of calibration to make it work at its best will be like 80 to 90% of headset mic. and the head set mic, I think, at it is best is somewhere in 80s% in average (of all users).
aside from the PC side there is the user who is a complex variable, accent, volume, pitch, current emotional state, all of these change what the computer hears
with current technology (MS SR) it is gonna be always limited until there is a system that simulate our brains to be more than 95% (even us we miss interpret the things we hear many times). even systems like Siri and Cortana ..etc. now they focus on natural language... but how we hear and how we react is much more complex ... i.e. we can tell if the speech is directed to use even if our name (prefix simulate that a bit, though we can back process even if it is mentioned at the end or in the middle, in fact someone might make it look like that the speech is directed to you but it is meant for someone else, and both we will understand, based on the tone) is not mentioned based on direction of the sound, how many people in the room, the tone, the person speaking among many other factors ... in one words; context processing which current computers are far from getting there.
all of this to say "with open mic, you have to set a very realistic goals and start with a very simple command system (i.e. having one way of saying things that are precise and not close to other commands)" that will affect your experience drastically ... and then add slowly more commands, one at a time ...
I second James, if you need help we need to know well your setup ....