SpeakerIdentApp - Text-Independent (TI) Speaker Identification Application -------------------------------------------------------------------------- COPYRIGHT --------- $Header: /cvsroot/marf/apps/SpeakerIdentApp/README,v 1.18 2012/01/09 04:03:22 mokhov Exp $ Copyright (C) 2002 - 2012 The MARF Research and Development Group <http://marf.sf.net> The application is distributed under the same terms as the MARF itself. In short: it's a BSD-style license. See COPYRIGHT for details. GENERALITIES ------------ NOTE: This demo application utilizes the MARF framework probably the fullest. You have to have the latest marf-<version>.jar file, which you can obtain from <http://marf.sf.net/>, and put it somewhere under your CLASSPATH, EXTDIRS (Java's extesions directory), or application directory itself. The -bin and -bundle distros have the .jar file within them, (but it may not be the latest). You will need JDK/JRE 1.4 or higher for it to function properly. This is (still) a console-based application. GUI is still being developed. For the fresh install you need to retrain the system before you can start feeding testing samples into it. If you work with our sample set, to save some time you may download our trainin-sets tarball with pre-trained data. RUNNING ------- (This applies to the (pre)compiled binary version; if you downloaded sources only, see the next section prior this one.) Aside from running from the Makefile as described in the follow up section, you can run testing and wrapper scripts or .jar/.class files directly. Linux SpeakerIdentApp and testing.sh are the shell scripts. The former is merely a wrapper for the Java .class or .jar files, which passes all the options to the Java application directly. The script tries first to locate a .jar and then a .class file in the current directory by default. The testing.sh default script is an 'all-out' testing with retraining for all possible combinations of algorithms. WARNING: it may run as long as 10 days depending on the hardware for all algorithms and voice samples, so if you want something for your quicker needs, adjust it accordingly by removing options you don't need. Windows SpeakerIdentApp.bat and testing.bat are Windows equivalents of the above shell scripts, so all of the above applies to them. Windows also has a 'retrain.lnk' shortcut to testing.bat that automatically supplies the --retrain option when double-clicking it. BUILDING FROM SOURCES --------------------- (This does not apply to you if you downloaded the binary-only distro.) UNIXen It is assumed you have a GNU-style make, which comes under some UNIXen under the name gmake. Type: $ make to build it $ make train-run to build, train the system, and then test $ make ident-run to just batch-test it w/o retraining $ make javadoc-html to make a javadoc-style reference to the application's code; the generated documents will be placed under the ./html subdirectory $ make clean to remove all .class files $ make maintainer-clean to also remove output, text, ppm, log, diff, and some other temp. developer files NOTE: You may need to tweak the Makefile for your specific preferences. Windows For JBuilder users there is the SpeakerIdentApp.jpx project file, you can use that if you are unfamiliar with the make utility or suffer from its unavailability. Eclipse For both Linux and Windows, you can import the Eclipse project files from the application directory, called .project and .classpath. DOCUMENTATION ------------- Aside from this README, ChangeLog, the javadoc reference, and code comments this application doesn't have any documentation within. It's documented, however, along with the MARF framework itself, in the MARF's report-manual. Grab its PDF version from either MARF's -doc distro or the our web page: <http://marf.sf.net>. CONTACT ------- Report bugs to: <marf-bugs@lists.sf.net> Comments, ideas, contributions, etc. direct to: <marf-devel@lists.sf.net> <mokhov@users.sf.net> EOF