Effects of Speaker Count, Duration, and Accent Diversity on Zero-Shot Accent Robustness in Low-Resource ASR

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yong, Zheng-Xin, Pratap, Vineel, Auli, Michael, Maillard, Jean
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916780424822784
author Yong, Zheng-Xin
Pratap, Vineel
Auli, Michael
Maillard, Jean
author_facet Yong, Zheng-Xin
Pratap, Vineel
Auli, Michael
Maillard, Jean
contents To build an automatic speech recognition (ASR) system that can serve everyone in the world, the ASR needs to be robust to a wide range of accents including unseen accents. We systematically study how three different variables in training data -- the number of speakers, the audio duration per each individual speaker, and the diversity of accents -- affect ASR robustness towards unseen accents in a low-resource training regime. We observe that for a fixed number of ASR training hours, it is more beneficial to increase the number of speakers (which means each speaker contributes less) than the number of hours contributed per speaker. We also observe that more speakers enables ASR performance gains from scaling number of hours. Surprisingly, we observe minimal benefits to prioritizing speakers with different accents when the number of speakers is controlled. Our work suggests that practitioners should prioritize increasing the speaker count in ASR training data composition for new languages.
format Preprint
id arxiv_https___arxiv_org_abs_2506_04364
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Effects of Speaker Count, Duration, and Accent Diversity on Zero-Shot Accent Robustness in Low-Resource ASR
Yong, Zheng-Xin
Pratap, Vineel
Auli, Michael
Maillard, Jean
Computation and Language
Sound
Audio and Speech Processing
To build an automatic speech recognition (ASR) system that can serve everyone in the world, the ASR needs to be robust to a wide range of accents including unseen accents. We systematically study how three different variables in training data -- the number of speakers, the audio duration per each individual speaker, and the diversity of accents -- affect ASR robustness towards unseen accents in a low-resource training regime. We observe that for a fixed number of ASR training hours, it is more beneficial to increase the number of speakers (which means each speaker contributes less) than the number of hours contributed per speaker. We also observe that more speakers enables ASR performance gains from scaling number of hours. Surprisingly, we observe minimal benefits to prioritizing speakers with different accents when the number of speakers is controlled. Our work suggests that practitioners should prioritize increasing the speaker count in ASR training data composition for new languages.
title Effects of Speaker Count, Duration, and Accent Diversity on Zero-Shot Accent Robustness in Low-Resource ASR
topic Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2506.04364