Reproduce “Phoneme-aware Encoding for Prefix-tree-based Contextual ASR” 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
Hayato Futami, Emiru Tsunoo, Yosuke Kashiwagi, Hiroaki Ogawa, Siddhant Arora, Shinji Watanabe
Futami et al. introduce phoneme-aware encoding for prefix-tree-based contextual Automatic Speech Recognition, which integrates pronunciation information with textual representations to improve rare word recognition. The method reduced Word Error Rate by approximately 5-10% relative for English and 8-15% relative for Japanese compared to grapheme-only encoding.
Key insights
- Incorporating phoneme information helps the model better align speech input with biasing words, especially for words with pronunciations that differ from their orthographic forms.
- The method's effectiveness across both English and Japanese, languages with different writing systems, suggests broad cross-linguistic applicability.
- Leveraging pronunciation information directly through phoneme embeddings and a phoneme-aware query improves generalization, particularly for unseen words with similar phonetic patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.