The RemoveVocals Audio Team
The engineers and researchers who design, train and ship the audio models behind every tool on this site.
Last reviewed: April 10, 2026Léa Moreau
Lead ML Audio EngineerLéa leads the machine-learning work behind the Vocal Remover, Stem Splitter and Audio-to-Lyrics tools. She specializes in spectrogram-domain neural source separation and runs the pipeline that ports research-grade models down to in-browser inference without measurable quality loss. Eight years working on music information retrieval, previously in audio R&D at a streaming-platform research lab.
Julien Bertrand
DSP EngineerJulien owns the classical DSP side of the toolkit: the phase-vocoder pitch changer, the time-stretcher, the spectral-subtraction noise reducer, the chromagram-based Key Finder and the onset-detection BPM Finder. Former audio-plugin developer with a decade of experience writing real-time DSP in C++ and, more recently, in WebAssembly and AudioWorklets.
Camille Rousseau
Mastering & QA LeadCamille runs the listening-panel and mastering work. Every model update is validated against a reference set of commercial tracks spanning pop, rock, hip-hop, electronic, classical and spoken word. She also tunes the AI Mastering tool's loudness, true-peak and tonal-balance targets to match EBU R128 and streaming-platform norms. Background in mastering engineering and psychoacoustics.
How we review tool pages
Every tool page on RemoveVocals carries a visible "last updated" badge and lists a reviewer. We re-run the page through our editorial checklist at least once per quarter: we re-test the underlying model against a fresh sample set, update the accuracy statistics if they have moved, refresh any competitor comparison, and bump the dateModified field in the structured data. Any significant change also triggers a short changelog entry on the blog.
Editorial standards
We don't run sponsored comparisons, we don't take payment for placement on any listicle, and we don't publish a tool page until it has passed a blind A/B listening test against at least two commercial alternatives. When we claim a statistic — accuracy, speed, SDR score — it comes from an internal benchmark you can reproduce by uploading the same test file yourself.
Contact
Reach the team via the About page. For legal, licensing or press enquiries see the Legal Notice.