Reproduce “Does per-frame early exit pay? A compute-matched study of dynamic depth for on-device speech enhancement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Clément Laroche, Riccardo Miccini
The authors train a causal speech enhancer with exits at multiple depths, then fine-tune only the exit heads to encourage deeper outputs not to worsen, enabling fixed-depth models or per-frame routing. On 16-kHz VoiceBank–DEMAND, routing reaches 2.719 PESQ at 8.48 MMAC/s, but repaired exits differ by less than 0.01 PESQ; hardware results favor within-ladder flexibility, not a universal dynamic advantage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.