Georgi Dzhambazov's Projects
making a few web apps for testing
Lyrics-to-audio-alignement system. Based on Machine Learning Algorithms: Hidden Markov Models with Viterbi forced alignment. The alignment is explicitly aware of durations of musical notes. The phonetic model are classified with MLP Deep Neural Network.
Scripts for computing common lyrics-to-audio alignment evaluation metrics. Usable evaluation for any token-based alignment (e.g. if token is word, phrase, note, section etc.) User for the evaluation of the MIREX Lyrics-to-audio challenge
A heuristic approach to the detection of choruses in vocal cover versions. Done at the WiMIR workshop at ISMIR 2019 https://docs.google.com/presentation/d/1WYXxChgo8DI_NknyndPdcOuxlhAL_428Olg2fWaVy6U/edit#slide=id.g6f4f88d309_0_57
My Docker scripts and Dockerfile for several frameworks.
matlab prototype for drum transcription sytem described in the paper https://drive.google.com/file/d/0B4bIMgQlCAuqdGVRbVNNbzJfeUU/view
The Dunya music browser. Developed using Django 2
scripts to create mapping from English phoneme models as feed forward network multilayer perceptron network onto a GMM turksih phoneme model
the dataset used in the paper https://drive.google.com/file/d/0B4bIMgQlCAuqdGVRbVNNbzJfeUU/view
C++ library of algorithms to extract features from audio files, including Python bindings.
Python Hidden Markov Models framework. Adapted for computationally optimal Viterbi forced alignment. Added Explicit Duration model
Parses models created by the HTK Toolkit (http://htk.eng.cam.ac.uk/) as text files into Python class. It enables then various operations with the models like visualization and comparison.
Example of the inverse MFCC essentia feature
smart-phone game app that teaches a technique to memorise music intervals. check out demo video:
Lyrics-to-Audio Alignment for Jingju Arias
singing voice with annotations of vocal onsets, based on the matched MIDI from http://colinraffel.com/projects/lmd/
lyrics-to-audio-alignement system. Initially done using HTK for rapid prototyping
matlab version of lyrics2audio using DTW. experimental
Python audio and music signal processing library. This is a fork adding support for synchronous tracking of vocal note onsets and metrical position in bar. The model used is Dynamic Bayesian Networks.
acapella recordings of Makam
Reproduce the htk-type of MFCC features using the essentia framework. The MFCC extracted with essentia are compared to these extracted with htk these extracted with librosa
Music Structure Analysis Framework
Examples of extracting acoustic features with essentia
Tools to match MB to other datasets/collections (like echonest)
Manual annotations of audio segments that correspond to sections from score with singing voice present
PDNN: A Python Toolkit for Deep Learning. http://www.cs.cmu.edu/~ymiao/pdnntk.html
The data needed to generate my phd thesis