Improvements in Language Modeling, Voice Activity Detection, and Lexicon in OpenASR21 Low Resource Languages

Gupta, Vishwa; Boulianne, Gilles

doi:10.1007/978-3-031-48312-7_6

Cited by 2 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For contexts where matching language models are not readily available, more realistic results are to be expected, such as in Coto-Solano et al (2022) where median word error rate (WER) ranged from 18-66% for Cook Islands Maori. Even with language models, WERs remained high for low-resource languages: 32.91% for read speech in Bemba language in Sikasote and Anastasopoulos (2022) and 48% for Kurmanji Kurdish in Gupta and Boulianne (2022).…”

Section: Introductionmentioning

confidence: 99%

Automated speech recognition of Indonesian-English language lessons on YouTube using transfer learning

Maxwell-Smith,

Foley

2023

Proceedings of the Second Workshop on NLP Applications to Field Linguistics

View full text Add to dashboard Cite

Experiments to fine-tune large multilingual models with limited data from a specific domain or setting has potential to improve automatic speech recognition (ASR) outcomes. This paper reports on the use of the Elpis ASR pipeline to fine-tune two pre-trained base models, Wav2Vec2-XLSR-53 and Wav2Vec2-Large-XLSR-Indonesian, with various mixes of data from 3 YouTube channels teaching Indonesian with English as the language of instruction. We discuss our results inferring new lesson audio (22-46% word error rate) in the context of speeding data collection in diverse and specialised settings. This study is an example of how ASR can be used to accelerate natural language research, expanding ethically sourced data in low-resource settings.

show abstract

Section: Introductionmentioning

confidence: 99%

Automated speech recognition of Indonesian-English language lessons on YouTube using transfer learning

Maxwell-Smith,

Foley

2023

Proceedings of the Second Workshop on NLP Applications to Field Linguistics

View full text Add to dashboard Cite

show abstract

Advances in OpenASR21 Evaluation with Increased Temporal Resolution for Speech Self-supervised Learning Models

Gupta

2024

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Improvements in Language Modeling, Voice Activity Detection, and Lexicon in OpenASR21 Low Resource Languages

Cited by 2 publications

References 17 publications

Automated speech recognition of Indonesian-English language lessons on YouTube using transfer learning

Automated speech recognition of Indonesian-English language lessons on YouTube using transfer learning

Advances in OpenASR21 Evaluation with Increased Temporal Resolution for Speech Self-supervised Learning Models

Contact Info

Product

Resources

About