Vosk and numbers recognition
Does anyone know about vosk and numbers recognition?
Everything is working fine with words like
"audio on"
"looping on"
"record"
"clear track" (only when armed so only need one of those)
But when I speak numbers
"arm 1" - > "arm one" or "arm won" works
arm 2 -> "arm do"
arm 3 -> "arm tree"
arm 4 -> arm for"
arm 5 -> ???
arm 6 -> ???
arm 7 ->???
arm 8 -> "arm ate"
you can see that pattern
//
I also need dynamic situation
"tempo 60"
"tempo 70" , 80 90 all work (kind of)
anything over 100 getting nothing or weir results , 120 -> I get 21
nothing for 71 72 73 82 88 etc
I can deal with base 10 results for tempo- and just add a bit that say faster, or slower , [incdec]
//
But will have many situations where I need to say the number - especially 1-10
//
Here is the python so far -
this all works- very well - im impressed - now I can play my bass and just say 'record' without having to stop playing - pick up pencil- push record (4 beat count-in) put down pencil - try to play
, but eventually I will want to change meter on the fly
say things like 3 beats 4 measure or 5 beats 3 measures (3/4 x4 measures, 5/4x3 measures)
its sounds crazy - but real good solution to yelling at the computer and actually making it useful!
Thank for assistance.
if you run something like that locally the model is so small that it can not grasp the context of a sentence which only has two words ("arm ate")
i bet it also does not work too well to arm a, b and c instead.
"first, second, third" would be a better attempt.
have you tried yet to limit vosk to a list of allowed words?
Hi thanks. I didn’t think about first second third etc. willl try that.
I did make a number list and a correction for some numbers - if 21 then 120 ( something like that) but decided it was a waste of time because I would have a big giant list of possible corrections.
if you run something like that locally the model is so small that it can not grasp the context of a sentence which only has two words ("arm ate")I tested vosk and whisper. Whisper is for paragraphs, vosk is for short word groupings.
As I mentioned it works great with non-numerical words. Yes I tried a b c - but that become a problem- a bee sea
But it does bring up interesting concept phrases sound funny live
“Ate arm for lunch”
“One arm race”
“Arm bee sting”
Arm sea swim”
Arm for sale”
:)
I’ll check
Arm five hands
Arm six fingers
Arm seven brides
//
I’m using the one Bang phaser a lot these days in the phrase “one Bang one “
“one Bang two” “one bang three” work great
//
I’m also thinking , lento, largo, adagio, allegro , presto etc to bang tempo, and use [incdec] with faster /slower- similar to ‘hey metronome’
//
Finally- all my tracks can play an independent meter perfectly in sync so I need to come up with a way to say things like “arm one” “3 beats” “4 measures “”record”
— “arm 2” 5 beats” “6 measures” “record”
Etc like that
Or even more complex, I want to say "arm one" "four beats" "three measures" "tempo 72" "record"
"Arm two" "seven beats" "five measures" "tempo 96" "record"
Things like that
allegro adagio presto... i like that. :) reminds me on "sudo -cleanmykitchen".
yes, trying to correct the input will become a mess. if anything works, then it is using a positive list. of course you had to add new words to the json for python whenever you add something in max in this... maybe not too elegant.
Beethoven wrote a lot of presto - super fast and hard to play!
My clean kitchen ->allegro -> subito lento -> tacet
ad impossibilia nemo tenetur