Gender Classification & Pitch Analysis

Hi folks!

As promised last week I now added some speech processing stuff to the blog. There is a gender classification tool where you can use gaussian distribution models to let the PC tell you whether speech data is from a male or a female speaker. Therefore a real tiny and easy machine learning strategy is used (training and testdata included).

The pitch analysis is done using linear prediction models. So you can find the formants within speech. There is also a short document included in the archive, wehere you can read through some testing results so you might understand these things easier and without any experiences in this field.

Rock on, Florian

How to practically get started developping a computer game in 3D?

Hi again, as promised, I added some new material to the publication section including the source code for a tiny open GL game. So if you are interested in developing computer games this example covers most of the things need to get started. You require the devil engine for that project which is available for free on the web – you just have to google. For the animation of the model, the CAL3D plugin for maja has been used, so might also need this library, but it is also an open source project. The archive contains the executables, so if you are just interested in playing the game, you might just launch the Demo.exe. Instructions on how to play the game and how to control the player just have a look to the readme.txt. You can switch between different shadow models for the characters (textured, projected) and there are some animation paths that use splines (linear, bezier, catmullrom), which can also be displayed. Projected shadows will propably require a fast machine because there are about 50 npc’s with shadows rendered in real time and without quality restrictions, so this requires really much cpu power.

I also added my bachelor thesis to the blog which is about gesture recognition using machine learning which I already have mentioned in an earlier post. I am afraid this paper is restricted to people that speak German but if you have questions about it, feel free to contact me any time.

This week I will propably upload the promised speech processing examples like the pitch detector or a speaker identification model to seperate male from female voices.

Improved MPEG Compression

Hi folks,

it has been some time since the last time I published some new stuff, but now I am back again. In the publications section you will find a Matlab example that implements MPEG2 compression for images. In addition the coding process has been improved and enhanced by block processing so the quality after compression is even better than compression for common MPEG2 compression. MPEG2 normally does not use block processing and applies a single weight matrix onto the whole image. Because I use block processing, the quality is better at the same error rate.

Recently, I am about to finish a tiny computer game that will be online soon. This might be interesting for those of you that like to start programming computer games. It implements already published examples like splines, AI, shadows and animation under OpenGL.