Anyone have speech / voice recognition working on OSX 10.6?
Been trying sphinx2pd python script without much luck.
Most solutions I find are Linux based...
Using extended 0.42.5
Thanks! mark
On 11-06-24 02:48 AM, mark edward grimm wrote:
Anyone have speech / voice recognition working on OSX 10.6?
Maybe you can use the built-in speech recognition (i know nothing about OSX)?
Been trying sphinx2pd python script without much luck.
I did this quick hack awhile ago and only tested on Linux.
Most solutions I find are Linux based...
I am so happy to read that :)
Using extended 0.42.5
I think it's better to use a voice recognition software (not a pd external) and then make the bridge between the software and pd (with shell, python, lua, OSC, emulation of keyboard / mouse, etc).
For kiku, i'm using Julius voice recognition (can be compile on OS X) and the Voxforge english acoustic model. The next version have Open Sound Control (client only) built-in. So maybe you can use VirtualBox inside OS X install Ubuntu and use kiku OSC (but what an ugly solution).
http://www.workinprogress.ca/kiku/about
à +
Em 24/06/11 04:31, patrick escreveu:
For kiku, i'm using Julius voice recognition (can be compile on OS X) and the Voxforge english acoustic model. The next version have Open Sound Control (client only) built-in. So maybe you can use VirtualBox inside OS X install Ubuntu and use kiku OSC (but what an ugly solution).
Patrick... nice job!
But.. how do you recognize so many languages? (I'm confused, as you say you are using an english model and the "online dictation" page recognizes other languages as well...)
josé
Hi José,
But.. how do you recognize so many languages? (I'm confused, as you say you are using an english model and the "online dictation" page recognizes other languages as well...)
kiku is only able to recognize English (15k words) and Japanese (60k words). The web page use HTML5 "x-webkit-speech" on Google Chrome and it's a complement to kiku since the accuracy of the English acoustic model is not yet ready for dictation. so kiku = single word recognition and the web page a complement for dictation.
http://www.workinprogress.ca/KIKU/dictation.php
à +
hey Patrick.... just trying Kiku out. Works for a minute then stops responding. specifically when i open pd-extended.
maybe its my setup?
osx 10.6 -> vmware fusion -> ubuntu 11.04 -> kiku 0.1 + pd-extended 0.42.5
On Fri, Jun 24, 2011 at 7:52 AM, patrick puredata@11h11.com wrote:
Hi José,
But.. how do you recognize so many languages? (I'm confused, as you say you are using an english model and the "online dictation" page recognizes other languages as well...)
kiku is only able to recognize English (15k words) and Japanese (60k words). The web page use HTML5 "x-webkit-speech" on Google Chrome and it's a complement to kiku since the accuracy of the English acoustic model is not yet ready for dictation. so kiku = single word recognition and the web page a complement for dictation.
http://www.workinprogress.ca/KIKU/dictation.php
à+
Pd-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
hi mark,
i am surprise that it's working on ubuntu 11.04 - kiku is not showing in the taskbar (a bug in compiz), but it's not affecting the recognition afaik.
as for your problem, kiku have an option to recognize only the words if, in your example, pd-extended is opened causing kiku to restart his engine (thus portaudio). that might be your problem (i cannot reproduce this problem here, ubuntu 10.10).
you change this behavior in Configuration: Dictionary: Match V2C (uncheck)
also you need to know that kiku is single word recognition. meaning if you speak a sentence it will output a single word (with a very bad score).
remove ~/.kiku folder to "reset" the installation.
pat
yeah.... i figured it out. pd just needs to be started from the terminal. then all works well!
also you need to know that kiku is single word recognition. meaning if you speak a sentence it will output a single word (with a very bad score).
yeah figured that.... which I can make works for what im doing (i think) although full dictation or semi-dictation would be more ideal....
i am surprise that it's working on ubuntu 11.04
seems to work!!! idk....
cheers mark
On Sun, Jun 26, 2011 at 1:13 PM, patrick puredata@11h11.com wrote:
hi mark,
i am surprise that it's working on ubuntu 11.04 - kiku is not showing in the taskbar (a bug in compiz), but it's not affecting the recognition afaik.
as for your problem, kiku have an option to recognize only the words if, in your example, pd-extended is opened causing kiku to restart his engine (thus portaudio). that might be your problem (i cannot reproduce this problem here, ubuntu 10.10).
you change this behavior in Configuration: Dictionary: Match V2C (uncheck)
also you need to know that kiku is single word recognition. meaning if you speak a sentence it will output a single word (with a very bad score).
remove ~/.kiku folder to "reset" the installation.
pat
hi mark,
yeah figured that.... which I can make works for what im doing (i think) although full dictation or semi-dictation would be more ideal....
it is possible to write a grammar for Julius / kiku, see the documentation:
http://julius.sourceforge.jp/en_index.php?q=en_grammar.html (using FILLER, in your case, will be useful)
then you just need to change ~/kiku/language/julius.conf
remove -w dictionary add -dfa grammar.dfa (you have to create this file) -v dict (already included with the English language)
this will allow dictation in kiku, but since the English acoustic model is not yet ready for dictation it might not work that well. anyone who have some time and is a native English speaker, please contribute to voxforge:
à +
hello,
thanks for response!
the apple built in speech recognition is for "Speakable Items, built into Mac OS X and located in the Speech pane of System Preferences," ... basic words to control computer...
I think im looking for more of "speech dictation" for recognizing longer phrases.\
I have a project coming up and I am just weighing my options. Basically I am monitoring radio signals via a police scanner and want to pick out key words. It doesn't even have to be that accurate really. just so certain words can trigger instance in PD....
cheers mark
On Fri, Jun 24, 2011 at 3:31 AM, patrick puredata@11h11.com wrote:
On 11-06-24 02:48 AM, mark edward grimm wrote:
Anyone have speech / voice recognition working on OSX 10.6?
Maybe you can use the built-in speech recognition (i know nothing about OSX)?
Been trying sphinx2pd python script without much luck.
I did this quick hack awhile ago and only tested on Linux.
Most solutions I find are Linux based...
I am so happy to read that :)
Using extended 0.42.5
I think it's better to use a voice recognition software (not a pd external) and then make the bridge between the software and pd (with shell, python, lua, OSC, emulation of keyboard / mouse, etc).
For kiku, i'm using Julius voice recognition (can be compile on OS X) and the Voxforge english acoustic model. The next version have Open Sound Control (client only) built-in. So maybe you can use VirtualBox inside OS X install Ubuntu and use kiku OSC (but what an ugly solution).
http://www.workinprogress.ca/kiku/about
à+
Pd-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-
mark edward grimm -
padovani -
patrick