Our research addresses the challenge of speech recognition for African accented speech by exploring the linguistic relationships of accent embeddings obtained through AccentFold. Our exploratory analysis of AccentFold provides insights into the spatial relationships between accents. The findings reveal that accent embeddings group together based on geographic and language family similarities, capturing phonological and morphological regularities specific to language families.
Our experimental setup demonstrates the practicality of AccentFold as an accent subset selection method for adapting ASR models to targeted accents. With a WER improvement of 3.5%, AccentFold proves promising for improving ASR performance on accented speech, particularly for African accents, overcoming data scarcity and budget constraints.
While our research has shown positive results, one limitation is the use of a single pre-trained model for fine-tuning in our experiments. Although this model demonstrated promising performance, future research could explore a broader range of pre-trained models to enhance the applicability and effectiveness of AccentFold in diverse speech recognition contexts.
[
Collection
]
[
|
...
]