Reproduce “ViSpeechFormer: A Phonemic Approach for Vietnamese Automatic Speech Recognition” 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
Khoa Anh Nguyen, Long Minh Hoang, Nghia Hieu Nguyen, Luan Thanh Nguyen, Ngan Luu-Thuy Nguyen
ViSpeechFormer introduces an Automatic Speech Recognition (ASR) framework for Vietnamese that explicitly models phonemic representations, leveraging the language's phonetic orthography and stable syllable structure. This approach achieves state-of-the-art performance on Vietnamese ASR benchmarks, demonstrating superior generalization to out-of-vocabulary words and reduced training bias compared to character and subword-level baselines.
Key insights
- Explicitly modeling the phonemic structure of Vietnamese, specifically the initial, rhyme, and tone components, effectively leverages the language's phonetic orthography and stable syllable structure for ASR.
- A linguistically-informed phoneme-based vocabulary leads to a compact decoder and allows the model to build words compositionally, reducing reliance on memorizing frequent lexical items.
- Phoneme-level decoding significantly enhances generalization capabilities, particularly for out-of-vocabulary words, and mitigates training bias inherent in character- and subword-based models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.