Abstract
Session 2: Machine Learning
This study applies Automatic Speech Recognition (ASR) to a sociolinguistic corpus of Australian English. We compare a human transcription of excerpts from 20 urban and regional speakers with a transcription generated by Microsoft’s Azure AI Speech. The Word Error Rate is comparable to previous studies, and is not impacted by the sociolinguistic variables of speaker region and gender, nor the phonetic variable of vowel formants. Despite the overall low rate of transcription errors, our findings suggest that the quality of certain vowel categories that are particularly characteristic of Australian English can impact on the accuracy of the ASR-generated transcription.
This study applies Automatic Speech Recognition (ASR) to a sociolinguistic corpus of Australian English. We compare a human transcription of excerpts from 20 urban and regional speakers with a transcription generated by Microsoft’s Azure AI Speech. The Word Error Rate is comparable to previous studies, and is not impacted by the sociolinguistic variables of speaker region and gender, nor the phonetic variable of vowel formants. Despite the overall low rate of transcription errors, our findings suggest that the quality of certain vowel categories that are particularly characteristic of Australian English can impact on the accuracy of the ASR-generated transcription.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of the Nineteenth Australasian International Conference on Speech Science and Technology |
| Editors | Olga Maxwell, Rikke Bundgaard-Nielsen |
| Publisher | Australian Speech Science and Technology Association Inc |
| Pages | 27-31 |
| Publication status | Published - 2024 |
| Event | 19th Australasian International Conference on Speech Science and Technology - University of Melbourne, Melbourne, Australia Duration: 3 Dec 2024 → 5 Dec 2024 https://assta.org/sst-2024/ |
Publication series
| Name | Proceedings of the Australasian International Conference on Speech Science and Technology |
|---|---|
| ISSN (Electronic) | 2207-1296 |
Conference
| Conference | 19th Australasian International Conference on Speech Science and Technology |
|---|---|
| Abbreviated title | SST2024 |
| Country/Territory | Australia |
| City | Melbourne |
| Period | 3/12/24 → 5/12/24 |
| Internet address |