Past Meetup

Topological Data Analysis & Speech Prosody Modelling

This Meetup is past

79 people went

Faculty of Technology and Metallurgy, Amphitheatre

Rudgjer Boshkovikj 16 · Skopje

How to find us

Amphitheatre Faculty of Technology and Metallurgy, Rudgjer Boshkovikj 16 · Skopje

Location image of event venue

Details

Dear members of DataScience Macedonia,

We are very excited to announce the next MeetUp, that will be held on[masked], Thursday, from 18:00 to 20:00 at the Amphitheater of TMF (Faculty of Technology and Metallurgy).

------
The first talk will be about Topological Data Analysis, given by Marko Karbevski.

Topology is the part of mathematics that is primarily concerned with shapes. Topological Data Analysis is an emerging field within data science that helps you find topological and geometric features within your data that could give meaningful insight (through, say, clustering or visualisations), and provide features for your classification algorithm or reduce the dimensionality of your data.
The presentation will present some of the techniques in this field and their application to a case study. The following papers are suggested for more information:
https://arxiv.org/pdf/1609.08227.pdf
https://arxiv.org/pdf/1710.04019.pdf

Bio: Marko Karbevski is a Data Scientist at Sorsix. He graduated in mathematics before moving to the Data Science field. His modest experience includes predictions concerning social media (via locally linear modelization), transportation (using NNs) as well as data analysis for clients. Also, he is an intern at MANU where he explores the world of topological data analysis.

-------

The second talk will about Data Science in Speech, given by Dr. Branislav Gerazov.

Speech encodes linguistic, paralinguistic and non-linguistic information via its prosody (intonation, energy and rhythm). How it does this is still an open issue. Within the framework of the ProsoDeep project (https://gerazov.github.io/prosodeep/) we built upon the modelling paradigm of the Superposition of Functional Contours (SFC) model by incorporating deep learning methods, to both decompose prosody into its constituent contours, and also capture a part of their variance. The Variational Prosody Model (VPM) comprises a network of variational encoding (recurrent) neural network contour generators, which map the linguistic context of the contours into a prosodic latent space. The prosodic latent space can then be used to analyse prosodic phenomena, as well as to generate prosodic contours in a speech synthesis (TTS).

Bio: Branislav Gerazov finished his bachelor, master and PhD studies at FEEIT in 2007, 2011 and 2014. He is an assistant professor at FEEIT since 2015, where he has taught the courses of Electroacoustics, Digital Audio Systems, Biomedical Engineering and Biomedical Electronics. He has participated in 3 international projects, in 2 of which as principal investigator. In his scientific work he has published more than 70 papers in journals and conferences. His interests are the digital analysis, processing, synthesis and recognition of audio signals, specifically speech, as well as biomedical signals.

---------

Free snacks and drinks! Provided by Sorsix, Macedonia!