With Honors

Today, the Magnus Master’s Thesis defense has been held and the project has been qualified with an “Excellent With Honors”.

Defense presentation

The defense of the Magnus Master’s Thesis will be held tomorrow at Enginyeria i Arquitectura La Salle. For this event, the presentation that will be used is available in the forge under the documentation section, along with the thesis released about a month and a half ago.

The frames present a theoric basis of the Automatic Speech Recognition and Speech Enhancement techniques used in the project, a block diagram description of Magnus, the adaptation of the original speech recognition engine (Sphinx-4) to Catalan and the performance results extracted from the application.

Magnus understands Spanish

According to the proposal that alfem posted on January the 30th., a translation of Magnus into Spanish, now it has come true. It is basically the same code as the original 0.1.1 release (in Catalan) but aimed at the Spanish language.

In the forge, a binary  distribution named “magnus-spanish” has been made available, as well as a simple copy-paste source-code patch that works along with the 0.1.1 release of Magnus in order to enable the continuity and improvement of the project, but now in Spanish too.

Magnus 0.1.1 released

This new release of Magnus includes a speech enhancement module that increases the robustness of the recognition system (for more details please refer to the Master’s Thesis) and a couple of new spoken commands to:

• put Magnus in “sleeping mode” so that it doesn’t bother the user while doing something else.
• change the focus of the application thus emulating the Alt+Tab keyboard shortcut.

Magnus release 0.1.1 is available in the forge as a source code distribution (including all the extra materials pertaining to the research developed) and a binary distribution (pre-compiled with Java version “1.6.0_06″).

Master’s Thesis released

Yesterday 29 September in the afternoon, the Master’s Thesis was handed in La Salle Engineering University. In about a month’s time the thesis defence will be held.

The thesis is publicly available in atrilla (see “Projects and Publications”), in the forge or directly downloaded at the following link:

http://www.salle.url.edu/~st12809/data/files/magnus_master_thesis.pdf

Speech enhancement analysis and development with Scilab

The speech enhancement unit for Magnus is under development using the open source platform for numerical computation Scilab.

As stated in the first release of the impovement guidelines the starting point is a speech enhancement system proposed by Shankar Chanda and Park in the ICASSP’07 meeting. Now the system baseline has been programmed with Scilab, however, it is still being improved in order to achieve the best results for the application.

The Scilab scripts are available through SVN with the command:

svn checkout https://forja.rediris.es/svn/csl2-magnus/scilabworks

Once the algorithmic side of the problem is properly adjusted the next step will be the translation of the Scilab code to the Java code and thus the inclusion of the speech enhancement unit into Magnus.

Development code available through Subversion (SVN)

The present development code for Magnus is available through the source code management system Subversion. The repository is placed in the forge and is accessible anonimously with the command:

svn checkout https://forja.rediris.es/svn/csl2-magnus/trunk

The repository can also be browsed with WebSVN and ViewCVS. Please check the forge to get the appropriate links.

Magnus through Java Web Start

On the Links section there is a link to execute Magnus through the Java Web Start (JWS) technology. JWS will automatically check against the server that the users have the latest release, downloading the necessary newer files in case they need them, and launching the application afterwards.

JWS makes easy the process of  installation since it makes it transparent to the users who wish to quickly run Magnus. It is ideal for having a try on the application.

Improvement guidelines (release 1)

In the Digital Signal Processing field:

• Speech Enhancement based on [Shankar Chanda, P. and Park, S., "Speech intelligibility enhancement using tunable equalization filter", ICASSP, 2007].

In the IT field:

• Client-server architecture migration.
• Language portability through Java I18N.
• Visual interface (SWING library instead of AWT).
• Java Web Start available.

Magnus 0.1.0

The first release of Magnus is available on the forge of the project.

It is still a beta version, many improvements are yet to be done, but to date, under noiseless or low noise environments, the program responds fairly well. There is though a high dependence on the ambient conditions and the quality of the microphone. This is the main flaw that will be prioritized for the next release. A better speech intelligibility is the goal to be obtained through signal processing techniques.