Exploring the world of coding, one project at a time.

A gallery of my personal projects.

Audio sleep stage classification based on Traditional and Deep features.

This project develops a non-invasive method to classify sleep stages—awake vs. asleep—using audio recordings. By analyzing respiratory sounds from the PSG-Audio dataset, two approaches were compared: traditional feature extraction with a Support Vector Machine (SVM) and a deep learning model using Convolutional Neural Networks (CNN).

...

The SVM approach offers rapid predictions with 89% accuracy, ideal for real-time applications. Meanwhile, the CNN model provides a higher accuracy (93%) but requires more processing time.

View more

Bulgarian Text-to-Speech

I have trained a Bulgarian Text-to-Speech (TTS) model using the Variational Inference (VITS) architecture. This project purpose is to generate high-quality synthetic speech from input text in the Bulgarian language. Hear audio samples of the artificial voice below.

"Аз съм завладяващ глас, говорещ от сърцето на дигиталните технологии. Симфония от нули и единици, бродирани в гоблен от безупречен звук. Стремя се да хипнотизирам и пленявам, оставяйки незаличима следа върху слуховите ви сетива."



"От новинарските организации най-често се очаква да се стремят към обективност, а самите репортери твърдят, че се опитват да покрият всички страни на даден въпрос безпристрастно."



"Първите обитатели на териториите, днес заети от Португалия, са били неандерталци и хомо сапиенс."

In the future, i plan to further train and release the model and add multiple speakers and languages.

View more

Slime Mold Simulation

Visualization of the Dijkstra"s algorithm in a grid-based environment. Every step is displayed by highlighting visited cells, simulating a slime-like growth. The visualization tries to imitate the movement of Physarum polycephalum.

View Video
...

Song sales database in MySQL

A university course project that demonstrates various SQL queries and functions.

View pdf