magcil.github.io · NCSR "Demokritos", Athens, Greece
We are a research group specialising in multimodal deep learning with a focus on audio, speech, and music analysis. Our work spans self-supervised representation learning, generative modeling, robustness evaluation, and applied AI systems. From fundamental research to deployed tools.
- Audio & Speech Analysis — classification, segmentation, feature learning, emotion and behavioral signal analysis
- Music Information Retrieval — song identification, fingerprinting, genre/mood analysis, music generation
- Robustness & Security — adversarial attacks on audio models, noise robustness, deepfake detection
- Few-Shot & Self-Supervised Learning — low-resource audio classification, SSL backbone evaluation
- Applied AI Systems — real-time inference pipelines, cloud APIs, on-premise deployments
| Repository | Description |
|---|---|
| deepaudio-x | Python library for training deep audio classifiers using SSL backbones (BEATs, PaSST, MobileNet) with minimal code |
| pretrained-ssl-audio-backbones | Collection of open-source self-supervised audio models for downstream tasks |
| deep-audio-fingerprinting-benchmark | CNN-based neural audio fingerprinting with contrastive learning and Faiss indexing for robust song identification |
| audio-adversarial-attacks | Black-box adversarial attack evaluation on state-of-the-art audio transformers (BEATs, PaSST, AST) using PSO and Differential Evolution |
| audio-few-shot-learning | Few-shot audio classification combining prototypical networks and contrastive learning across the MetaAudio benchmark |
| song_mine_project_public | End-to-end song mining system: deep fingerprinting, Faiss-based retrieval, and interactive demo |
| public-speakers-quality | Dataset and baselines for automatic public speaking quality assessment (387 annotated conference presentations) |
| e-prevention-challenge | Code and resources for the e-Prevention challenge on mental health monitoring (2nd Position as a lab in ICASSP2025) |
📧 tyiannak@gmail.com · 🌐 magcil.github.io · 📍 IIT, NCSR Demokritos, Athens