Cross-lingual Embedding Clustering for Hierarchical Softmax in Low-Resource Multilingual Speech Recognition

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yang, Zhengdong, Liu, Qianying, Li, Sheng, Cheng, Fei, Chu, Chenhui
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915127903649792
author Yang, Zhengdong
Liu, Qianying
Li, Sheng
Cheng, Fei
Chu, Chenhui
author_facet Yang, Zhengdong
Liu, Qianying
Li, Sheng
Cheng, Fei
Chu, Chenhui
contents We present a novel approach centered on the decoding stage of Automatic Speech Recognition (ASR) that enhances multilingual performance, especially for low-resource languages. It utilizes a cross-lingual embedding clustering method to construct a hierarchical Softmax (H-Softmax) decoder, which enables similar tokens across different languages to share similar decoder representations. It addresses the limitations of the previous Huffman-based H-Softmax method, which relied on shallow features in token similarity assessments. Through experiments on a downsampled dataset of 15 languages, we demonstrate the effectiveness of our approach in improving low-resource multilingual ASR accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2501_17615
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cross-lingual Embedding Clustering for Hierarchical Softmax in Low-Resource Multilingual Speech Recognition
Yang, Zhengdong
Liu, Qianying
Li, Sheng
Cheng, Fei
Chu, Chenhui
Computation and Language
Sound
Audio and Speech Processing
We present a novel approach centered on the decoding stage of Automatic Speech Recognition (ASR) that enhances multilingual performance, especially for low-resource languages. It utilizes a cross-lingual embedding clustering method to construct a hierarchical Softmax (H-Softmax) decoder, which enables similar tokens across different languages to share similar decoder representations. It addresses the limitations of the previous Huffman-based H-Softmax method, which relied on shallow features in token similarity assessments. Through experiments on a downsampled dataset of 15 languages, we demonstrate the effectiveness of our approach in improving low-resource multilingual ASR accuracy.
title Cross-lingual Embedding Clustering for Hierarchical Softmax in Low-Resource Multilingual Speech Recognition
topic Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2501.17615