record, compare and recognize

Feb 13, 2009 at 9:43pm

record, compare and recognize

Hi
I’m trying to make simple voice commands to control video.
Have searched through the forum and found that the story is rather old and complicated. Is not there somewhere in Max an object, which could record “command” and then compare received phrase to it? smthng like “GO” and “STOP”.
I don’t need “voice recognition”, but think about maximum 20 commands, which a machine can be trained to recognize.
I’m on windows, so aka.listen~ does not work …
thanks for attention!
alex

#42287
Feb 13, 2009 at 10:28pm

I hooked MacSpeech Dictate up to work with Max, obviously this is speech recognition software and mac-only but there are PC equivalents, though they tend to be quite pricey. I’m not aware of any other externals or in-max solutions but maxobjects.com might throw something up.

lh

#151315
Feb 14, 2009 at 1:46am

Quote: Onandoff wrote on Fri, 13 February 2009 15:43
—————————————————-
> Hi
> I’m trying to make simple voice commands to control video.
> Have searched through the forum and found that the story is rather old and complicated. Is not there somewhere in Max an object, which could record “command” and then compare received phrase to it? smthng like “GO” and “STOP”.
> I don’t need “voice recognition”, but think about maximum 20 commands, which a machine can be trained to recognize.
> I’m on windows, so aka.listen~ does not work …
> thanks for attention!
> alex
—————————————————-

If you can fork out $100 you can get Dragon Naturally Speaking, now in version 10 (get the basic version as that’s all that you’d need). Not sure if there’s a demo version, there was of version 8 that let you run it 5 times before paying. I really liked it and bought version 8. With some training (really only about a half hour) its accuracy was excellent. It’s a bit processor-heavy but that was on a considerably older machine, and is probably not an issue now.

So, with that running in the background, it can insert voice-recognized speech into any text app or field, including the Textedit object in Max. Send periodic bangs to report what it’s gotten, route the words you want, and voila. Works great and there’s pretty much no limit to the number of words you can match. Use a [coll] for matching if you want big lists, and note that certain short words (without a sentence context) are often actually tougher for it to recognize, whereas several-syllable ones are generally always on.

It’s actually kind of creepy magic to see it putting the text on there, getting some words wrong in the sentence, then retroactively fixing them as it recognizes the unfolding grammatical/meaning context… Hard to wrap my brain about how tough that would be to program, let alone the voice-matching algorithms…

There likely are cheaper alternatives, but this one is relatively affordable and works very well, and can really be a help with lots of other computer tasks (not just typing, but surfing is better too) if you get the hang of using it. There are a lot of clever additional niceties included which aren’t purely “speech-to-text”, they’re commands to (say) an open browser to go Back, New Tab, Save, etc.

CJ

#151316
Mar 15, 2009 at 1:50am

I have a similar question that I’ve been trying to fix on my own with no success. I can route out the word “text” from the textedit box, but after that, I guess textedit sends out the entire contents as one thing. I want to have it so the user-entered input has each word read separately, so that certain “key” words will trigger other events. “Stop” is one example. I thought maybe I could replace the spaces between words to be line breaks so that the textedit would then only read each word one line at a time, but I can’t get that to work, either. I’m very new to max; can anyone help?

#151317
Mar 15, 2009 at 1:58am

If you’re using a mac then have a look at this recent thread where I explained how to use the OSX’s built in voice recognition utility to trigger keypresses which you can read in max. I hope it’s helpful and sorry for all the spelling errors!

http://tinyurl.com/speechrec

lh

#151318
Mar 15, 2009 at 10:12pm

Sio wrote on Sat, 14 March 2009 20:50I have a similar question that I’ve been trying to fix on my own with no success. I can route out the word “text” from the textedit box, but after that, I guess textedit sends out the entire contents as one thing. I want to have it so the user-entered input has each word read separately, so that certain “key” words will trigger other events. “Stop” is one example. I thought maybe I could replace the spaces between words to be line breaks so that the textedit would then only read each word one line at a time, but I can’t get that to work, either. I’m very new to max; can anyone help?

Try this, it uses [iter] which may be what you need.

– Pasted Max Patch, click to expand. –
#151319

You must be logged in to reply to this topic.