Maxime Baelde Avatar

Maxime Baelde, Ph.D

Senior Data Scientist | Machine Learning Engineer @ Antvoice
Ph.D. in Artificial Intelligence
Music Producer @ MB Prod
Building Duetly

About Me

I’m a Senior Data Scientist, with some experiences in digital audio processing, retail and digital marketing. I manage projects from start to finish, mainly: data analysis, model experimentation, training and serving.

My Ph.D. thesis was focused on real-time audio source localization, identification and separation. My supervisors were Christophe Biernacki (Scientific Deputy at Inria Lille) and Raphaël Greff (R&D Director at A-Volute).

I graduated from University Lille 1 (Ph.D. in Applied Mathematics, Master Research Applied Mathematics) and Ecole Centrale de Lille (Master Decision and Data Analysis).

I’m also a self-taught musician (guitarist, bassist, pianist) for about 10 years, mainly playing Progressive Metal musics (Meshuggah, Periphery, …) and Muse. I have a YouTube channel where I post guitar and bass covers of various songs. I also have a Bandcamp page for my solo music project Fracton. If needed you can stream my music from almost any platform like Spotify.

I’m also building Duetly, a revision and file-transfer platform for audio engineers and their artists, under the MB Prod brand.

Finally, I’m a casual video game player, Nintendaddict (The Legend of Zelda series) and Switch player.

You can find my résumé here.

Skills & Experience

Machine Learning & AI

  • • Deep Learning (TensorFlow, PyTorch)
  • • Generative Models
  • • Reinforcement Learning (Stable Baselines3, Gymnasium)
  • • Signal Processing

MLOps & Cloud (GCP)

  • • Airflow, Docker
  • • BigQuery
  • • Cloud Storage
  • • Vertex AI

Software Engineering

  • • Python (Advanced)
  • • SQL
  • • API Development (FastAPI)
  • • Git, CI/CD

Teaching & Academia

  • • Statistics & Mathematics
  • • Machine Learning Tutoring
  • • Academic Writing

Intellectual Property

  • • Patent Research & Review
  • • Patent Writing

Business Contexts & Apply Domains

Digital Marketing & Retail

  • • Real-Time Bidding Models
  • • Sales Forecasting

Audio Engineering

  • • Consumer Electronics Audio Processing
  • • Industrial Audio Applications

Publications

2026

Maxime Baelde. “From “Different” to “Better”: What It Would Take to Show That a Hiring Protocol Works”. Zenodo preprint, CC BY 4.0.
[doi]
Maxime Baelde. “Threshold-Cascaded Lexicographic Rewards for Multi-Constrained RL in Real-Time Bidding”. Submitted to TMLR (Transactions on Machine Learning Research), under review.
[openreview]

2024

Maxime Baelde, Nathan Souviraà-Labastie and Raphaël Greff. “Apparatus and method for audio analysis”. US Patent US11956616B2, April 9, 2024.
[patent]

2023

Maxime Baelde, Laurent Mareuge and Julien Roland. “Système de reconnaissance et d’identification de sources sonores en temps réel”. FR Patent FR3109458B1, August 26, 2022.
[patent]

2019

Maxime Baelde. “Modèles génératifs pour la classification et la séparation de sources sonores en temps-réel”. Ph.D thesis, Université de Lille, 2019.
[thesis] [code]
Maxime Baelde, Christophe Biernacki and Raphaël Greff. “Real-Time monophonic and polyphonic audio classification from power spectra”. In : Pattern Recognition 92 (august 2019), p. 82-92.
[paper]
Nathan Souviraà-Labastie, Maxime Baelde, Thomas Malet and Raphaël Greff. “Impact des conditions d’attaques sur les contre-mesures pour la reconnaissance du locuteur”. pre-print Gretsi.
[paper]
Maxime Baelde, Nathan Souviraà-Labastie and Raphaël Greff. “Influence of the attack conditions on countermeasures for Automatic Speaker Verification”. pre-print Interspeech.
[paper]

2017

Maxime Baelde, Christophe Biernacki and Raphaël Greff. “A mixture model-based real-time audio sources classification method”. 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 5-9 Mar 2017, pp. 2427-2431, New Orleans, USA.
[paper] [poster]
Maxime Baelde, Christophe Biernacki, Raphaël Greff. “Classification de signaux audio en temps-réel par un modèle de mélanges d’histogrammes”. 49èmes Journées de Statistiques, June 2017, Avignon, France.
[paper]
“Real-time source sources identification”. Presentation at 7èmes Rencontres Jeunes Statisticiens, 5 April 2017, Porquerolles, France.
“Real-time source sources identification”. Presentation at Modal’s Day on the first year of the thesis, 20 January 2017, Le Touquet, France.

2016

“Real-time identification, localization and separation of audio sources in multichannel audio streams, at ReSci : Acoustique, musique, et production sonore humaine ou animale”. Presentation at ANRT on the first 6 month of the thesis, 7 July 2016, Paris, France.
“Modélisation de spectres sonores par mélange de gaussiennes - Application à la reconnaissance de sources sonores”. Presentation for MASSICCC demo, 7 April 2016, Villeneuve d’Ascq, France.

2015

Maxime Baelde. “Modelization of HRTF filters and optimization for 3D audio effect”. Master thesis, 2015, Université de Lille.
[paper]

2014

Maxime Baelde, Jessica De Saedeleer et Jean-Pierre Hermand. “Experiment to evaluate musical qualities of violin strings”. In : The Journal of the Acoustical Society of America 136.4 (2014), p. 2284-2285.
[abstract]

Writing

Non-academic articles on engineering, machine learning and audio. Academic work is under Publications.

2026

Carnet de voyage : un an de reinforcement learning en production
July 2026
A field diary of one year running a reinforcement learning project for real-time bidding: building a market simulator from scratch, sculpting a reward function that encodes a business strategy, and an A/B test that invalidated a core latency assumption within hours. Ends on the reward-shaping inversion that turned out to be a theorem, now under review at TMLR. In French.
[LinkedIn (FR)] [paper]
4 self-hosting failures that return success
July 2026
Four ways a self-hosted server reports success while being broken: a 200 on a gated WebDAV endpoint, a healthcheck with no curl in the image, a backup that exits 0 with an empty archive, and a Wake-on-LAN packet that quietly evaporates. Companion repo: sober-selfhost.
[dev.to (EN)] [LinkedIn (FR)]

Code

Here you will find some code I developped, available on Github.

churn-poc

End-to-end churn prediction project: model training to API exposition with FastAPI.

[GitHub]

dadabovic-led-rgb-matrix

Scripts to drive RGB LED panels for live shows.

[GitHub]

generative-audio-source-models

Generative models for real-time audio source classification and separation (PhD thesis, Univ. Lille 2019).

[GitHub]

launchpad

2x2 soundboard Android app, built for my kids.

[GitHub]

nongaussian-mixtures

Scikit-learn-compatible mixture models (Dirichlet, Beta, binned Gaussian) from my thesis appendices.

[GitHub] [PyPI]

simple-vad

Voice Activity Detector: signal-processing beamforming + CNN classifier.

[GitHub]

sober-selfhost

Energy-sober self-hosting on consumer hardware: on-demand containers (Sablier) gated behind Caddy, multi-DB encrypted backups (Postgres/MariaDB/SQLite -> rclone crypt), bidirectional Wake-on-LAN over Tailscale, and battle-tested Docker healthcheck fixes.

[GitHub]

vst-plugins

A set of VST plugins

[GitHub]

Open Source Contributions

linkedin-mcp-server

Open-source MCP server for LinkedIn. Give Claude and any MCP-compatible AI agent access to profiles, companies, jobs, and messages.

[GitHub] [PR]

Music

I have my own mixing / mastering engineer business called MB Prod.

Here you will find some music I produced with my solo music project Fracton.

My Productions (Fracton)

Engineering Work (MB Prod)

  • GROINCK (mastering), available July 4th, 2026.