Implement an Autoregressive Model for Anonymized Voice-Synthesis at a Defense Vendor
Overview
What this challenge is about.
You receive a public-domain speech dataset (LibriTTS subset, around 50 speakers) and a fixed evaluation protocol (speaker-identifiability AUC, emotion-preservation MOS proxy, intonation correlation). Implement an autoregressive voice-anonymization model (Tacotron-style with a target-speaker embedding swap is fine for the budget) and run it against an off-the-shelf voice-conversion baseline. Report metrics with confidence intervals, document compute cost, and write a 4-page research note.
The Brief
What you'll do, and what you'll demonstrate.
Compare an autoregressive voice-anonymization model to an off-the-shelf baseline on speaker-identifiability, emotion preservation, and intonation correlation.
Earning criteria — what you'll demonstrate
- Implement autoregressive sequence models for speech
- Design fair benchmarks that hold protocol constant across systems
- Report results with confidence intervals and compute cost
- Communicate research findings to a non-public audience succinctly
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Research Scientist
Speech-generation research with formal benchmarking and confidence intervals is the standard early-career research scientist deliverable.
This challenge sharpens
- autoregressive-models
- speech-synthesis
- benchmarking
ML Researcher
Implementing autoregressive sequence models from a paper baseline is core ML research work at any speech or language group.
This challenge sharpens
- autoregressive-models
- voice-conversion
- pytorch
AI Safety Researcher
Voice-anonymization research at a defense-grade evaluation bar overlaps directly with AI safety research on privacy and identifiability.
This challenge sharpens
- voice-conversion
- evaluation
- benchmarking