Profile
My expertise is in speech technology, particularly speech synthesis and recognition. I am also versed in audio and biomedical signal processing and in the broader fields of signal processing, machine learning, and deep learning. I am enthusiastic about doing research and practical implementation of advanced signal processing and machine learning pipelines, as well as exchanging know-how with students and collaborators.
Highlights
Research
- 110 scientific papers and 2 books in speech technology (TTS, ASR, prosody modelling, speaker recognition, voice forensics and transformation), and DSP, ML & DL in audio and biomedical signal processing
- published in leading journals including IEEE TASLP, Speech Communication, and JASA, and flagship conferences, such as Interspeech
- led and participated in 7 international and 4 national research projects
- MSCA post-doc at GIPSA-lab, Grenoble-INP, France
- research collaboration with McGill University, Canada, UCL, UK, Idiap, Switzerland, IBEB, Portugal
Industry
- co-founder and ASR consultant at Speech AI, Skopje, Macedonia
- served as CTO at ListenUp Technologies, Tel Aviv, Israel
- R&D consultant for IBH-IMPEX, Germany, ReTell, UK, and Macedonian Gordian Technologies, ITEK Systems, Analitika, and Sentice
- co-authored a speech processing patent
Data
- prompt design and speech corpus recording and transcription
- speech and audio data analysis, denoising and post-processing
- a varied language portfolio including low-resource langugaes: English, French, German, Turkish, Mandarin, Vietnamese, and Macedonian
- work with challenging audio data for forensics
Teaching
- mentored 6 Master Students and more than 30 Bachelor students
- mentored student team that won the IEEE Signal Processing Cup 2023 at ICASSP
- mentored three student teams that won international Hardware&Software competitions
- authored teaching materials for 3 courses
Organization
- organized 90 talks and events as IEEE SP/EMB Chapter Chair
- served as Technical Program Chair at IWSSIP 2023
- organized WG and MC meeting as WG co-leader in COST Action a-STEP
- helped organize 10 conferences
- helped organize 6 student competitions, training schools and events
Code
- ProsoBeast-Annotation-Tool - web app for annotating prosody patterns
- evoc-learn - codebases for simulating vocal exploration with VocalTractLab
- Govorko - assistive technology Android app for AAC
- ProsoDeep - a PyTorch prosody model
- ollama-chat.nvim - Neovim plugin for chatting with Ollama LLM models
- contributed to coqui-ai/TTS
Experience
Associate (2019) and Assistant professor (2015) teaching Digital Audio Processing, Electroacoustics, and Biomedical Electronics.
Deep learning based modelling of English intonation.
Modelling speech prosody using deep learning under the ProsoDeep project funded by the EU Horizon 2020 under the Marie Skłodowska-Curie Actions (MSCA).
Work on Medical Microwave Imaging using Deep Learning.
Heading the development of algorithms for speech rate modification.
Development of speech processing algorithms involving machine learning.
Teaching courses in audio processing, electroacoustics, embedded systems, digital and analog electronics.
Teaching courses in audio processing, electroacoustics, embedded systems, digital and analog electronics.
Signal processing algorithm development involving machine learning.
Projects
Advancing Machine Learning in Vocational Education, PI, funded by Erasmus+.
High quality simulation of early vocal learning through neural models and articulatory synthesis.
Research project focused on the development of deep variational models of speech intonation. Project funded by Horizon 2020, Marie Skłodowska-Curie Individual Fellowship.
Industrial research project focused on the deployment of speaker recognition in call recording systems.
Research project on physiologically plausible intonation modelling. Project funded by Swiss National Science Foundation (SNSF).
Education
Thesis title: Noise Robust Automatic Speech Recognition for the Use in Systems for Vocal Interaction
Thesis title: Speech Synthesis System for Macedonian Based on Quasi-Diphones
Thesis title: The preservation of sound
Skills
Signal Processing
Machine Learning
Deep Learning
Natural Language Processing
Python
Scientific
Numpy, SciPy, Pandas
ML & DL
scikit-learn, PyTorch, Keras
NLP
NLTK, SpaCy
Visualisation
Matplotlib, Seaborn, Bokeh, Plotly
Web
Pelican, Flask, Django
General
JS, React
HTML, CSS
LLMs
Git
Linux, Bash, Zsh
Neovim, Lua
Languages
Macedonian and other Slavic
English
German
French
Interests
- Traditional music
- Mountain hiking
- Swimming
- Books on personal and professional growth