KVoiceControl for KDE1 ported (-ish, WIP) from OSS to libpulse.


Original README:

********************************************************************************
*****                              KVoiceControl                           *****
********************************************************************************

KVoiceControl is a tool that gives you voice control over your unix commands.


# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # 

There is now a mailing list for my little program: 

  [email protected]

To subscribe to the list send an e-mail to [email protected]
where the e-mail must contain the following in the mail body (NOT the subject) on a
single line:

subscribe kvoicecontrol <my e-mail address>

To unsubscribe use

unsubscribe kvoicecontrol <my e-mail address>

For information about the list

info kvoicecontrol

----------------------------------------------------------------------------------
This list is generously sponsored by Schreiber Online (http://schreiber-online.de)
Feel invited to visit their web site!! ;-)
----------------------------------------------------------------------------------

# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # 


------------------------------------------------------------------------------------
See the Help section within KVoiceControl for information on how to use the program.
------------------------------------------------------------------------------------


KVoiceControl uses a template matching based speaker dependent isolated word 
recognition system with non-linear time normalization (DTW).

Note, that isolated word does NOT necessarily mean ONE SINGLE word. It just describes a 
class of recognition systems among

 isolated word
 connected words
 continuous speech
 spontaneous speech

Consider the following example: 
-------------------------------

 KVoiceControl knows five utterances (being connected to appropriate commands ...)
  - connect to internet
  - netscape communicator
  - one xterm please
  - launch emacs NOT VI!
  - how are you today?
 
 Here the recognition vocabulary consists of five "words" (=utterances) that can 
 only be recognized one at a time, i.e. you cannot connect the words to build 
 "sentences" like
  <how are you today?> please launch <netscape communicator> and <connect to internet>

IMPORTANT NOTE:
---------------

As mentioned above you do not need to use one single word as an utterance ...
and it is strongly recommended to use LONGER SEQUENCES of words !!!!!
This is due to the base concept of this recognition system: template matching.
So if you used short commands like say <xterm> and <xedit> confusability would 
be increased significantly and therefore recognition accuracy drops rapidly!!!


INSTALLATION:
-------------

Standard should do: configure, make, make install


TECHNICAL ISSUES:
-----------------

Sound is being record at 16kHz, 16 bit, mono. I guess, this should be supported 
by any standard sound card ?!

NOTE: To be able to record sound data (on a Linux system) one needs to introduce a 
      user group that is allowed to access the sound device "/dev/dsp". For that purpose 
      you need to add something like
       sound:x:101:root,daniel
      to /etc/group and add all users that should be allowed to access the sound card, 
      here these users are root (the system administration user) and daniel. Then change 
      the permissions and group properties of /dev/dsp0 
      (this can only be done as user "root" !)
        chgrp sound /dev/dsp0
	chmod g+r /dev/dsp0
      Now "ls -l /dev/dsp0" shoul output something like
	crw-rw--w-   1 root     sound     14,   3 Apr 29  1997 /dev/dsp0
      Then you must restart Windows to complete the installation ;-)

      For answers to many questions concerning Linux and sound have a look at the 
      Linux Sound-HOWTO which is available from
       ftp://metalab.unc.edu/pub/Linux/docs/HOWTO/Sound-HOWTO

NOTE: Be sure to select "Microphone in" and to set "mic in"-volume to an appropriate 
      value (use [x|k]mix). Then use Calibrate_Micro from the menu "Options" within 
      KVoiceControl to adjust the program to the microphone level

NOTE: Up to now, Linux's standard sound system (OSS free) does not support full duplex
      mode. That means you cannot play sound files when KVoiceControl is in action mode, 
      as the program maintains a reading socket connection to the audio device!
      This is not very satisfying as otherwise one could use a speech synthesis
      program (e.g. mbrola) to let the computer answer appropriately ... :-(
      That would be MULTIMEDIA !!! ;-)

      Options to circumvent this problem:
      - Reportedly there exists a patch for the free OSS that adds to it full duplex support
      - spend $20 for the commercial OSS available from www.opensound.com
      - Other sound systems for Linux support full duplex (e.g. ALSA - Advanced
        Linux Sound Architecure, residing at http://alsa.jcu.cz/)

      - A newly introduced option within KVoiceControl enables the user to have commands executed
        while detection mode is switched off, afterwards detection mode is turned on again.
        This is very convenient to produce synthesized speech as a control feedback!

Recognizer details:
-------------------

* isolated word recognition

* template matching using non-linear time normalization (DTW)
  - DTW using symmetric Sahoe&Chiba warping function
  - Adjustment window width: 40 (adjustable)
  - starting corner sloppiness: 6 
  - parallel warping planes and Branch&Bound comparison
  - nearest neighbour wins template matching (roughly)

* preprocessing:
  - Hamming window
  - Fast Hartley Transform (FFT) -> power spectrum
  - (log) melscale coefficients
  - mean substraction
  - IMPORTANT: Amplitude normalization is not yet performed on the wave data
               So it is important to
	        - keep about the same distance between microphone and mouth
                - speak at about a constant loudness level


================================================================================
                                     Last changed: Wed Nov  4 23:18:36 MET 1998
                                               Daniel Kiecza <[email protected]>