AI Vocal Stem Separation Quality Enhancer

Reltrix Studio

Remote · $100k – $130k / year · Full-time · Posted 4 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

Vocal stem separation — isolating the vocal track from a mixed music recording — is a foundational capability for remix production, karaoke generation, music licensing verification, and vocal isolation for hearing accessibility. The state of the art has improved dramatically over the past five years. What hasn't improved proportionally is the quality of artifacts that remain after separation: the bleed of instrument frequencies into the vocal stem, the timbral coloration the separation model introduces, and the handling of reverb and room sound that was mixed into the original before separation. We build AI music production tools for 180 studios and independent producers. Our current vocal separation pipeline uses an off-the-shelf open-source model. The gap is post-processing quality enhancement: taking the raw separation output and improving it through artifact removal, bleed reduction, and timbral restoration that makes the isolated vocal stem production-ready without manual touch-up from a mix engineer. You'll build the vocal stem quality enhancement pipeline: training artifact reduction models on paired degraded-clean vocal datasets, implementing frequency-domain bleed suppression, and building the perceptual quality evaluation that correlates model output scores with mix engineer quality ratings.

Fully remote.

Responsibilities

Train artifact reduction models on paired degraded vocal stems and clean reference recordings
Implement frequency-domain bleed suppression removing residual instrument frequencies from isolated vocal stems
Build timbral restoration models reducing the coloration that stem separation models introduce
Develop reverb tail handling that preserves natural vocal space while removing instrument reverb contamination
Create perceptual quality evaluation correlating automated scores with mix engineer quality assessments
Build enhancement pipeline integrating with existing separation workflow as a post-processing stage

Requirements

3+ years in audio ML or signal processing with music source separation, audio enhancement, or speech processing experience
Strong Python and PyTorch skills with spectrogram-based audio model development
Understanding of music mixing, frequency masking, and perceptual audio quality dimensions
Experience training models on paired audio datasets with perceptual quality objectives
Familiarity with mix engineering workflows and vocal production standards