Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

PARCO: Phoneme-Augmented Robust Contextual ASR via Contrastive Entity Disambiguation

작성자
  • Haebom

Author

Jiajun He, Naoki Sawada, Koichi Miyazaki, Tomoki Toda

Outline

This paper proposes Phoneme-Augmented Robust Contextual ASR via COntrastive Entity Disambiguation (PARCO) to address the challenges faced by automatic speech recognition (ASR) systems, which struggle with domain-specific named entities, particularly homonyms. PARCO integrates phoneme-aware encoding, contrastive entity disambiguation, entity-level supervision, and hierarchical entity filtering to improve speech discrimination, ensure complete entity detection, and reduce false positives under uncertainty. It achieves a character error rate (CER) of 4.22% on the Chinese AISHELL-1 dataset and a word error rate (WER) of 11.14% on the English DATA2 dataset under 1,000 distractors, significantly outperforming existing methods. It also demonstrates robust performance improvements on domain-specific datasets such as THCHS-30 and LibriSpeech.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel ASR model that effectively solves the homonym problem by utilizing phoneme-level information.
◦
We improved the accuracy and stability of object recognition through contrastive learning and hierarchical filtering.
◦
We have proven that our method performs better than existing methods on various datasets.
•
Limitations:
◦
There is a lack of analysis of the computational complexity and resource consumption of the proposed model.
◦
Further research is needed on generalization performance across different languages and domains.
◦
Further evaluation of robustness in real-world environments is needed.
👍
Slashpage로 제작됨