Learnable Fractional Superlets with a Spectro-Temporal Emotion Encoder for Speech Emotion Recognition
International Conference on Learning Representations (ICLR)
Research / Montréal
Alaa Nfissi
Learning from signals. Making models understandable.
Signal Press BureauSelected material ↙
A note in the margin
Research file
Cross-attentive emotion fusion
Studying the relationship between audio, text and video through cross-attention and feature fusion.
Follow this question →SIGNAL PRESS BUREAU / RESEARCH & PRACTICE
Concordia · TÉLUQ · CRISE
Research, software & human questions.
Alaa Nfissi is a doctoral candidate in artificial intelligence and information systems engineering at Concordia University in Montréal. His research connects learnable signal representations, speech and emotion analysis, interpretable models and reliable multimodal learning.
He is a research assistant at Université TÉLUQ and a research member of CRISE at UQAM. His doctoral research is supervised by Nizar Bouguila and co-supervised by Wassim Bouachir. He holds a research master’s degree from École Polytechnique de Tunisie and an engineering degree in computer science, specializing in data science, from ESPRIT.
Lines of inquiry
FILE 01
Research direction
Speech, emotion & human context
Learning from speech while keeping the speaker, the recording conditions and the limits of interpretation in view.
Open dossierFILE 02
Research direction
Learnable signal representations
Wavelets, wavelet packets and fractional superlets as adaptive representations for emotion analysis and speech enhancement.
Open dossierFILE 03
Research direction
Interpretable & multimodal learning
Understanding acoustic features and examining how audio, language and visual cues interact in emotion recognition.
Open dossierSelected publications
PAPER / 012026
Learnable Fractional Superlets with a Spectro-Temporal Emotion Encoder for Speech Emotion Recognition
International Conference on Learning Representations (ICLR)
Further papers in this selection
- 2025
IEEE Transactions on Affective Computing
SigWavNet: Learning Multiresolution Signal Wavelet Network for Speech Emotion Recognition
A trainable multiresolution wavelet network combines raw-speech representations with spatial and temporal attention.
- 2025
Applied Intelligence
From unaltered raw waveform to emotion: Synergizing convolutional and gated recurrent networks for holistic speech emotion analysis
A journal study of CNN-n-GRU on TESS, IEMOCAP and RAVDESS, with generalization to spontaneous speech left open.
Inside the research
FILE 01
active
Adaptive time-frequency learning
A research file connecting learned wavelets, wavelet packets, denoising and fractional superlets.
Open dossierFILE 02
active
Cross-attentive emotion fusion
Studying the relationship between audio, text and video through cross-attention and feature fusion.
Open dossierBeyond the research paper
Teaching & mentoring — Invited instruction at Concordia, professional workshops and student research.
Talks & public conversation — AQPS, La Presse, Radio-Canada and institutional coverage of the work.
Research software — Implementations, documentation and links back to the studies.
Recognition
IEEE International Conference on Semantic Computing
Best Student Paper Award · ICSC 2024
Awarded for Unlocking the Emotional States of High-Risk Suicide Callers through Speech Analysis, by Alaa Nfissi, Wassim Bouachir, Nizar Bouguila and Brian Mishara.
Source recordQuestions worth sharing.
For research conversations, collaboration and thoughtful exchange.