Neural Audio / DSP Engineer: Vocal Resynthesis & Watermark Robustness

We are seeking a highly specialized Neural Audio / DSP engineer or researcher for a confidential music-production R&D project involving generative audio, vocal reconstruction, audio provenance, watermark robustness, and high-fidelity resynthesis. This is not a conventional mixing/mastering job. We are specifically interested in candidates with meaningful experience in one or more of the following: • Digital signal processing (DSP) • Neural audio / generative audio • Audio watermarking and watermark robustness research • Audio provenance / fingerprinting systems • Generative-audio forensics • Neural audio restoration or resynthesis • Voice reconstruction / voice conversion • Source separation and artifact repair • Spectral, phase, and temporal analysis • Neural codecs, diffusion audio, or speech/music synthesis • Preservation of vocal timbre during reconstruction • Python / PyTorch / MATLAB audio research workflows PROJECT OBJECTIVE We have a proprietary vocal-production reference whose exact vocal identity, phrasing, timbre, microdynamics, formant character, timing, and emotional performance are extremely important. We want to investigate: 1. What measurable generative/provenance characteristics exist within the supplied audio. 2. How those characteristics respond to conventional editing, restoration, encoding, legitimate resynthesis, and reconstruction. 3. Whether portions of the signal can be reconstructed or resynthesized while maintaining extremely high perceptual similarity to the original vocal. 4. What sonic degradation or alteration occurs under different transformation methods. 5. Whether the defining sonic characteristics of the target vocal can ultimately be reproduced from new human vocal recordings. The primary goal is preservation, reconstruction, technical understanding, and development of a repeatable vocal-production process. We are not looking for generic EQ/mastering, a basic online AI detector, or someone whose entire solution consists of uploading the audio to a consumer “watermark removal” website. We want someone who genuinely understands what is happening inside the waveform. INITIAL PAID TEST The first engagement will use a short controlled test sample rather than an unreleased commercial master. We would expect: • Technical analysis of the supplied audio • Clear explanation of methodology • Before/after audio where appropriate • Objective measurements where possible • Assessment of vocal/timbre preservation • Explanation of what changed and why • Recommendation for a larger R&D workflow Strong performance on the initial test may lead to a larger 1–3 month research and production engagement. CONFIDENTIALITY Supplied material may not be published, shared, placed in a portfolio, or used for AI/model training without written permission. This project is intended for legitimate technical research, restoration, reconstruction, and music production. We will comply with applicable distributor/platform disclosure and provenance requirements. We are not seeking to falsify provenance or misrepresent required disclosures. TO APPLY Please answer the following: 1. Describe the most technically similar neural-audio, audio-watermarking, provenance, voice-reconstruction, or resynthesis work you have personally performed. 2. Have you worked with or reproduced any published audio-watermarking or watermark-robustness research? Please be specific. 3. If given a vocal whose exact timbre must survive reconstruction, how would you objectively determine whether the reconstructed version still preserves the source vocal identity? 4. What tools/frameworks would you use for spectral, phase, temporal, codec, and perceptual analysis? 5. Describe your experience with neural codecs, diffusion audio, source separation, voice conversion, speech synthesis, or generative music/audio systems. 6. Please provide relevant research, GitHub repositories, publications, technical demonstrations, or before/after examples if available. 7. What would you recommend as the smallest meaningful paid experiment before committing to a larger project? Important: At the very end of your proposal, include the words GHOST SIGNAL so we know you read the complete brief.