Ubuntu Wiki

cvoicecontrol project overview

• cvoicecontrol original home page http://www.kiecza.net/daniel/linux/index.html

• cvoicecontrol maps speech to unix commands
• you say 'star trek' and cvoicecontrol executes the defined unix command 'vlc /home/me/startrek/*.avi'
• as of 5/2010, no alternative package exists to map custom speech to custom unix commands
• gnome-voice-control hard-codes the speech commands as well as the unix commands.
  • i.e., with gnome-voice-control, changing 'star trek' to 'make it so' requires a src change
• cvoicecontrol defines the map (speech, unix-command) in configuration, and is specifically designed to update the map quickly and easily (cool!)

cvoicecontrol is currently not in service...

• Why cvoicecontrol is not used widely as of 5/2010:
  • cvoicecontrol was written in 2000 by Daniel Kiecza
  • bugs/problems have blocked potential users of cvoicecontrol for years
  • looks like no one has used it since at least 2007

Overview of cvoicecontrol architecture

• cvoicecontrol compiles three binaries:
  • microphone_config - one-time microphone setup, outputs config file
  • model_editor - manager of (speech, unix command) maps, each map stored in distinct *.cvc file
  • cvoicecontrol - actual voice control daemon, loads a single *.cvc file, usage ./cvoicecontrol my-voice-commands.cvc
• voice recognition works as follows:
  • model_editor stores four or more samples of your voice for the same voice command
    • i.e. to have "make it so" launch your startrek dvd, you will record "make it so" four times in model_editor
  • cvoicecontrol matches utterances with the voice samples you recorded in model_editor, and then executes the associated unix command
  • the whole voice recognition process relies on the microphone config file generated by microphone_config

Problems preventing release of cvoicecontrol-0.9alpha

• bug in microphone_config binary causes hard crash:
  • after finishing configuration, microphone_config writes a config file out. this fails.
  • in microphone_config.c:1155, the implementation constructing a valid path, "$HOME/.cvoiceconfig/config", to write out the file, causes a hard crash
  • temporary fix: comment this code out and write the config file out to /tmp
  • because microphone_config is run once-per-microphone, moving the config file manually from /tmp to "$HOME/.cvoiceconfig/" is ok-for-now
• depends on ncurses4. Ubuntu 10.04 ships with ncurses5
  • libncurses4 and libncurses4-dev install ok with libncurses5 present

  • possible issue with symlink, that may interfere with ncurses5: /usr/lib/libncurses.so -> /lib/libncurses.so.4

• microphone_config and model_editor binaries, both which use libncurses, work in the rxvt terminal. xterm and Eterm do not work at all.

Happy Path to building and getting cvoicecontrol working

• !!! NOTE: this does not include the aforementioned changes to microphone_config.c, therefore microphone_config will crash when it attempts to write config file !!!
• building:

  • get src http://www.kiecza.net/daniel/linux/index.html

  • install libncurses4 and libncurses4-dev (available from http debian package repository)
  • ./configure
  • edit cvoicecontrol-0.9alpha/cvoicecontrol/Makefile:
    • note, this is *not* the root Makefile located at cvoicecontrol-0.9alpha/Makefile
    • search LIBNCURSES,

    • old < LIBNCURSES =

    • new > LIBNCURSES = -lncurses

  • make
  • all three binaries should now be located in cvoicecontrol-0.9alpha/cvoicecontrol/
• voice commands setup:

  • see documentation for all stages of this process at the original project website http://www.kiecza.net/daniel/linux/index.html

  • install rxvt
  • run rxvt, run microphone_config
  • proceed through microphone configuration
    • Select Audio Device, /dev/dsp1 is generally your usb headset
    • Write Configuration, this step will fail and hard crash unless you make the changes to microphone_config.c mentioned above
  • exit microphone_config, mv the config file to $HOME/.cvoiceconfig/config
  • run model_editor, see original documentation mentioned above
  • in model_editor, Load/Save Speaker Model, be sure to include the .cvc extension in the file name
• running the cvoicecontrol daemon:
  • cvoicecontrol your-voice-map.cvc
  • at this point, mine "just worked". I spoke "play music" and my vlc unix command was executed. Note I had to speak very loudly because my generated microphone_config file tells cvoicecontrol to wait until I speak loudly :-P.

Runtime Gotchas

• if your unix command does not terminate immediately, i.e. if you launch vlc, you must postfix the fork & ampersand on the unix command when specifying it in model_editor. Failure to do this will cause cvoicecontrol to wait for the child process to terminate before continuing to process voice commands.

  • i.e. my voice command map contains ("play music", "/home/me/bin/music-start&"), and music-start is a script which launches vlc

• the speach matching is pretty liberal, and may (especially during initial configuration) accidentally execute the incorrect unix command. Only use transient read-only unix commands with cvoicecontrol (no rm : - ).

• model_editor requires absolute paths for unix commands. ~/bin/flashing-lights -> /home/me/bin/flashing-lights