• CV
• RESUME
• COMPUTER VISION
• NATURAL LANGUAGE PROCESSING
• DIGITAL SIGNAL PROCESSING
• DATA SCIENCE
• CONTACT

PRESENTATION

Skills to develop machine learning systems for signal processing applications. That is to say applications concerning sounds (audio), speech (speech), biological signals, images, etc.

SUMMARY

Signal Processing Techniques
Machine Learning for Signal Processing
Signal Processing Tasks
Signal Processing Applications
Signal Processing projects

SIGNAL PROCESSING TECHNIQUES

Signal processing is a multidisciplinary field that encompasses various techniques for analyzing, modifying, and extracting information from signals.

• Signal Acquisition
• Image, Video, biomedical signals
• Sampling and Quantization
• Filtering
• Detection and Estimation
• Time-domain Analysis

MACHINE LEARNING FOR SIGNAL PROCESSING

Machine learning techniques for signal processing vastly improve analysis, classification and prediction.

• Decompositions
• Classification
• Clustering
• Deep learning
• Optimization
• Time Series Models
• Generative models

SIGNAL PROCESSING TASKS

• Signal Analysis
• Signal Processing
• Speech Synthesis: Text-To-Speech (TTS),
• Speech Recognition: Emotion Recognition, Automatic speech recognition (ASR), Speech-to-Text (STT)
• Speech Generation: Text-To-Speech
• Voice Conversion: Speech-to-Speech (S2S)
• Speech Enhancement
• Speech Separation
• Speaker Diarization
• Speaker Verification
• Spoof Detection

SIGNAL PROCESSING APPLICATIONS

• Text-To-Speech (TTS) : Text processing + Speech synthesis
• Speech-to-Text (STT) : Speech recognition + Text processing
• Speech-to-Speech (S2S) : Speech recognition + Speech synthesis

SIGNAL PROCESSING PROJECTS

<

Voice cloning
Speech synthesis with conditioning

Programming language: Python
Libraries for Audio: Tacotron2, TorchAudio, LibRosa, Wave, SciPy, PyDub, IPython
Libraries for Text: TorchText, Text, SpeechRecognition, WordCloud
Libraries:: Pytorch, Torch, HuggingFace, Matplotlib, logging, Sys, OS, ...
Models: Tacotron2 and WaveGlow trained on the JLSpeech dataset
See the voice-cloning project.

• Tâche 1: Préparation et traitement des données
• Tâche 2: Modélisation avec Tacotron2
• Tâche 3: Évaluation et optimisation

Classification of bird songs by species
Signal processing
Analysis and Extraction of Spectrogram Image Information
Kaggle Competition: Birdcall Identification

Programming language: Python
Libraries: Pytorch, LibRosa and OpenCV
Model: CNN
See the Birdcall-identification project.

Generate a handwritten digits
Build a generative model for handwritten digits
Analysis and Extraction of Spectrogram Image Information

Programming language: Python
Libraries: Pytorch, LibRosa and OpenCV
Models: CNN
Datasets: CNN

• Tâche 1: Try learning a GMM
• Tâche 2: Improve these samples
• Train an autoencoder
• Train a GMM in the latent space
• Generate images