Delay between attention acknowledgement and command readiness

[DanWheeler]

There is a 1-2 second delay between the time the computer acknowledges attention and when it is ready for your command. Is there any way to shorten this? It makes "conversation" awkward especially for someone who is not used to it.

for example:
Me: "Computer"
Computer: "Yes?"
Me: "Turn off Amandas room floor lamp"

what the computer actually hears:
"das room floor lamp"

make sense? I could turn off the attention requirement but that's not such a good idea.

thanks,
dan

[sbessel]

What voices are you using?

[Skibum]

Perfect candidate for the Brightspeech voices!

[Rupp]

What's the speed of your pc?

[DanWheeler]

I have a 1.6Ghz P4 WinXP using (V5) Microsoft Mary. So I gather this is a fixable performance issue? Thanks SkiBum, I'll read up on BrightSpeech. Should 1.6 be fast enough? I just started eval of HS yesterday and I am very impressed.

[johnwpb]

Dan, Well being you are using MS Mary, this problem is probably not related to the voice you are using. MS Mary is a synthesized voice, and SHOULD be almost instant, and without the delays you mentioned.

It could be a memory issue, Hit Ctrl-alt-del and see how much memory is actually available when your system is loaded with all the normal things you are using. You can actually keep this open and isue a few voice commands and see the live processor usage and memory usage when you are spealing to the system. This delay could be because your memory is full, and using virtual memory from the hard drive, which would slow things up conciderably.

[Rupp]

Dan,
One warning about the BrightSpeech voices. They are currently not available for purchase and can only be demoed for 30 days. HS sells other voices that are equally as good that they sell that would amaze you if your using Microsoft Marry. Many members on this BB would argue about the merits of one voice over the other but I believe they all would agree that any add on voices are a lot better than the delivered Microsoft voices.

Back on your question. Does it happen every time the command is issued or every time? What X10 interface are you using?

[Skibum]

Agreed... You will find very different opinions on voices. The only common ground is that MS Mary is awful!

[DanWheeler]

MS Mary suits my needs for now - as long as I can understand her, thats all I need. I'm just incredibly impressed that my computer can understand *ME*!

As for the delay - I can't figure this thing out. Sometimes when I get Mary's attention, I can issue my command even before she's finished saying "Yes?" and the action carries out successfully. This seems to work much better for non-event-type, combined commands such as "Turn on - Amandas Room - Floor Light" But for event based commands I have MUCH better luck waiting until the status bar goes from "Not Listening, paused" to "listening for commands" 9 times out of 10 things work great as long as I wait until the status goes to "listening for commands"

I can't figure out why this is. I experimented with short words, long phrases, etc. and nothing seems to make a difference. I have plenty of memory (using 210 of 512) and processor usage is under 15%.

Funny thing happened - I accidentally got Mary into a command loop. I said "goodnight" then she said "goodnight" then she heard herself say "goodnight" and said "goodnight" to herself... and so on Gave me a laugh at least. I had to pull out an Eddie Murphy "LA LA LA LA LA LA LA LA LA LA" to get her out of it.

[johnwpb]

Dan, HAHHA! That would be very funny! The FIRST time it happened

Something is strange there, as HS, as you mentioned is supposed to go to not listening mode when the system is speaking. This prevents the loop that you encountered.

This seems to work much better for non-event-type, combined commands such as "Turn on - Amandas Room - Floor Light"

Are you using scripts to turn on the lights, or just set a individual device in the device properties for each light to turn on. The reason I ask, is if you have an event that runs a script that speaks, the computer will listen for the command, then stop listening if you have the script say something, and then listen again after the command, if the script speaks again, then it will go to not listen again, and then back to listen.

Take this small script, for example:

' (Home Seer is listening for incomming commands)
'You: Turn on the livingroom light
(This triggers the following script)
MyDevice = "A1"
'(Home Seer is NOT listening)
Hs.speak "ok"
'(Home Seer is Listening)
	hs.execx10 mydevice,"on",0
'(Home Seer is NOT listening)
Hs.Speak "I have turned on the livingroom lamp.  "
'(Home Seer is Listening)

This is probably not your case, but at least you can see that HS is not supposed to hear itself speak, so we know sowmthing is not right somewhere.

Your computer is plenty fast enough, and more than adequate memory, CPU usage is where it should be.... As this is not how HS is meant to function, You may want to file a help ticket explaining your problem, and making sure to to mention that HS actually got stuck in a loop listeing to itself

[rjh]

How many voice commands do you have? Is the delay only in a custom script? If the delay exists in the normal HS environment (not during a script), you can see how many voice commands you are creating by looking at the grm_command.xml file. This is the file that is given to SAPI to do the voice recognition. The larger this file is, the longer of a delay you will see. However, with the speed of your PC you would have to have quite a few voice commands. If you are using the Media Player plugin and have lots of MP3 files, that could slow it down.

[DanWheeler]

I believe I have 7 out of 9 total normal events that use voice commands. They play .wav files or turn on lights.

The delay I'm experiencing is slight, but enough to throw me off if I dont remember to wait.

I just noticed that if I keep talking after I say "computer" it will stay in the "not listening, paused" mode for as long as I keep talking.

then... i tried: (with an event setup to listen to "get me a beer")

me: "computer"
computer: "yes?"
me: (immediately) "blah blah blah wicky wicky wicky blah blah get me a drink blah blah blah blah get me a beer blah blah blah blah blah blah"

so... the status bar stayed in "not listening, paused" mode through the first set of blahs, through the wickys, through the "get me a drink", through the second set of blahs, and at the end of "get me a beer" it went to "listening for commands" before I could get to the third set of blahs.

Either something is really weird here or I dont understand how it works.

of course, this all may be irrelevant if the computer would just go to the "listening to commands" mode faster after the attention acknowledge phrase.

The grm_command.xml file is 7K / 210 lines. I dont think there's anything out of the ordinary. I've only been running the software for 3 days so I couldn't have done too much damage.

[DanWheeler]

it's possible i'm being way too picky here. The difference between when it works is so slight but it does mess me up. Here's an example of what I'm talking about:

http://danwheeler.gotdns.com/homeseer/getmeabeer.wma

i'm I just being too picky?

[rocco]

When you say it went to "listening to commands", are you refering to HomeSeer's status line? Keep in mind that, in Windows, updating the display is the lowest priority.

Is it possible that the system went 'listening' much sooner than the display indicated, possibly because it was too busy listening to update the display in a timely manner?

[johnwpb]

Another thought, you may be throwing off the voice command with your "wicky wicky's" (That does sound strange, sorry

When HS hears something, it will immediately begin to listen for commands.

For instance say the ONLY command you have HS set up to listen for, is "Get me a beer"

Now, you are having a conversation with someone in the room that goes something like this:

"I have had a really bad day at work, the boss was also having a bad day and I just wish someone would GET ME A BEER as that would just make me feel better. "

This entire phrase, from the very first word spoken, will take up quite a bit of CPU time, as the system is looking for something that it can understand. In most cases these CPU cycles are still going on long after you finish speaking, and needs to be completed before it can begin to listen again. If it is not EXACTLY sure what you asked for, This is where false triggers come into play as well. (EA: "Cry me a tear" could easily false trigger this event)

This can be complicated by having many phrases that HS understands being similar, such as:
"Turn on the over head light"
"Turn on the living room light"
"Turn on the bath room light"
"Turn on the Kitchen Light"
"Turn on the Hall Light"

They may a bit dissimilar, but simply put the ONLY differences in the above are just one small part of the entire phrase. The sounds "H". "L", "K" and "O" are just small differences to us, but much more to the computer when trying to compare 5 different phrases mathmatically, when 95% of each of the phrase is exactly the same. It takes that extra time and CPU cycles to do it's best to get the other 5% matched to do what we asked for.

Most of the time the VR is not that good to pick up your phrase out of the middle of that long sentence, but it does try, and devotes a lot of CPU cycles and time to that effort. In this example, you can at least see that it is doing its best to hear a phrase that it recognises. The longer the phrase, the longer the time it devotes looking for something that it can recognise.

I have a 3.2 GHZ machine with a gig of ram, and I still have to wait just that extra moment to continue a conversation with the computer.

After using the VR for a while, you will get to know just how long it does take for the VR to kick back in and listen again for a command after speaking. This is not a long delay, usually just slightly longer than conversation with another person, but there definately is a delay between HS saying something, and you being able to say something again when the system is actually ready to listen again.

Again, this normally is not a long delay, but you do have to give the system time to speak, and go back into listen mode again. Unfortunately, VR is "half-duplex" and your computer can not speak and listen for commands at the same time.

All this said, you may still have another problem, as you mentioned that your computer actually heard itself speak, and in turn responded to itself, causing an endless loop. This should NEVER be the case from what I have seen, as HS, by default, stops listening mode when it is speaking, to prevent this from happening in the first place. On another note, if your computer, using MS Mary as the voice , was able to trigger a VR command, you should at least do some additional training with the VR. That is of couse unless the person speaking actually sounds like Mary to begin with

Sorry, on a more serious note, Better clear and distinguishable voice matches with the VR, means less CPU cycles, and less time to process. If the phrase is not heard clearly, the CPU has to "replay what was heard" a couple times and come up with the best mathmetical match, and the best guess to match what was heard.


To quote a VERY WELL known phrase by everyone here for sure:

"Remember , the microphone hears everything, but sometimes it does not hear the person speaking very well!"

A good test:
Go to start/accessories/entertainment and Sound Recorder. Hit record and speak in the same normal tone that you would nomally when issuing commands. Play that back, and you will hear exactly what the computer hears. This will give you a good understanding why it takes that extra CPU to understand what you are saying. Most of the time the incomming audio is weak, or there is background noise that can interfere with the VR. This can be corrected by speaking closer to the mic that you have, or purchasing a better quality microphone.

As a last option, go to HS's "Tools" menu, and then to "Voice Recognition Properties..." Adjust the sliders in there to be either more accurate and slower, or less accurate and faster.

Just my 2 Cents...... errr.... $7.32 cents worth