Reproduce “SPEAR-Gen: Generation-Aware Pre-training for Unified Speech Representations” 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
Xiaoyu Yang, Arthur Hinsvark, Antonios Alexos, Osama Hanna, Philip C. Woodland, Yiting Lu
SPEAR-Gen trains a speech representation for both understanding and resynthesis by predicting fused, quantized targets from masked speech while learning mel-spectrogram reconstruction and residual flow matching. On SUPERB, its Base model improves final-layer ASR and emotion recognition over reproduced SPEAR; on LibriSpeech test-clean, PESQ-WB rises from 1.20 to 2.94 and speaker similarity from 0.467 to 0.892, not text-conditioned generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.