Sam Tureski's repositories
speechrecognition
My Bachelor's Thesis project, which involved German speech recognition using deep neural networks with transfer learning. Uses Keras interface for Tensorflow, scikit-learn, matplotlib, librosa.
audio-model
My Master's thesis project in audio classification using PyTorch and librosa. I achieve state of the art performance on the VGG-Sound dataset with the addition of a textual embedding layer to an existing dual-stream CNN framework.
vectors-webtool
Companion web app for "ViSpa (Vision Spaces): A computer-vision-based representation system for individual images and concept prototypes, with large-scale evaluation" (Guenther, F., Marelli, M., Tureski, S., & Petilli, M. A. ). A little webtool (at http://vispa.fritzguenther.de) for comparing image and word vectors, based on the Snaut website by Pawel Mandera
auditory-slow-fast-thesis
Implementation of "Slow-Fast Auditory Streams for Audio Recognition, ICASSP, 2021" in PyTorch
german-placenames
An interactive map project for German placenames
learning-python-through-NLP
manually implemented dependency parser and N-Gram-based language model; sentiment analysis with Keras
named-entity-recognition
Final project for Dr. Daniel de Kok's course on Deep Learning. We implement an RNN with Gated Recurrent Units in Tensorflow to achieve 95.56% accuracy on the CoNLL 2003 German dataset.
Neural-Network-Architecture-Diagrams
Diagrams for visualizing neural network architecture (Created with diagrams.net)
porcelluscavia.github.io
My personal website/portfolio built on a template by Hashir Shoaib
prof-names-counter
A tool for scraping and manipulating a large database of academic researchers.
Reinforcement-Learning-Hockey
Teaching an agent to play hockey by implementing Dueling Deep Q Networks in PyTorch
segment-modeler-DE
Using a neural network for prediction of German phones in audio files (Mel-frequency cepstral coefficients as features) in Tensorflow.
spectrogram-inversion
spectrogram inversion tools in PyTorch. Documentation: https://spectrogram-inversion.readthedocs.io
topic-classification-lda
Unsupervised learning (USL) techniques for topic classification on a parallel corpus. We use both K-means clustering and Latent Dirichlet Allocation (LDA).