Author Topic: Detailed recognition log output for recognized commands?  (Read 370 times)

SkyeWint

  • Newbie
  • *
  • Posts: 3
Detailed recognition log output for recognized commands?
« on: September 06, 2024, 03:39:11 AM »
When the speech engine hears unrecognized words, the unrecognized words are outputted to the recognition log, showing me what it heard. However, when there is a recognized command, it does not output a log of everything it heard, only the recognized command.

e.g.
Quote
Unrecognized : 'up to the envelope'
Quote
Recognized : 'ice' (confidence 85)

I used "ice" as an example here. That detection was when I said the word "spice" in continuous recognition mode. Ideally, is there a place where it logs all speech it is detecting, whether recognized or unrecognized? That way I can identify if it is hearing a command nested in a given word or not.

Thanks!

Pfeil

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 4782
  • RTFM
Re: Detailed recognition log output for recognized commands?
« Reply #1 on: September 06, 2024, 03:50:55 AM »
That is the output from the Microsoft speech recognition system. If a command is recognized, it returns the command phrase.

If you specifically require dictation output, use wildcard commands, which utilize the dictation capabilities of the speech recognition system, and as such can provide more information to the log.


The word/phrase "ice" is a very poor candidate for continuous recognition or wildcard commands though, being a very short, single-syllable word that will occur in other words, as you mention.

Continuous recognition and wildcard commands work best with longer, distinct, and unique phrases.

SkyeWint

  • Newbie
  • *
  • Posts: 3
Re: Detailed recognition log output for recognized commands?
« Reply #2 on: September 06, 2024, 11:48:14 AM »
Thanks, I recognize that "ice" is a poor candidate for continuous recognition for normal commands, but let's put that aside and say I want to do it anyway.

I've also attempted to use a wildcard such as * in the "When I say" block, but it doesn't seem to detect any words at all in that case. In the end, I don't really need the system to recognize everything with a wildcard, I just need some type of "recognized" output to display all the text it heard instead of the single word.

e.g. if I say "Seems like today was a good day", and my command word is "today", I'm looking for some kind of log of the whole sentence, and not for it to only say "Recognized: today" without displaying the rest of the sentence.

Pfeil

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 4782
  • RTFM
Re: Detailed recognition log output for recognized commands?
« Reply #3 on: September 06, 2024, 11:51:20 AM »
As mentioned, the Microsoft speech recognition system does not provide that information when recognizing commands, thus there is nothing more that can be shown in the log.

SkyeWint

  • Newbie
  • *
  • Posts: 3
Re: Detailed recognition log output for recognized commands?
« Reply #4 on: September 06, 2024, 12:01:29 PM »
Okay, so the answer is simply "this is not possible". That's unfortunate. I'll see if I can find another way to go about it, then. Thanks.

SemlerPDX

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Upstanding Lunatic
    • My AVCS Homepage
Re: Detailed recognition log output for recognized commands?
« Reply #5 on: September 06, 2024, 03:19:45 PM »
When the speech engine hears unrecognized words, the unrecognized words are outputted to the recognition log, showing me what it heard. However, when there is a recognized command, it does not output a log of everything it heard, only the recognized command.

I'm not sure what you mean by this.  Following your example from subsequent replies here, if I create a full wildcard command which only consists of "today" (surrounded by asterisks), and say the phrase, "Seems like today was a good day", the VoiceAttack Event Log shows the entire recognized phrase including the wildcard keyword, with a note of what word(s) contained in this sentence triggered a command in a profile:




To me, this seems like it is outputting everything it heard, and not just the actual command (word) which is present in the VA profile.

Is this not your experience?

To be clear, you will get this functionality using the wildcards method suggested by Pfeil above, and not merely using the 'Continuous Speech' recognition option(s).