LAHAJA: A Robust Multi-accent Benchmark for Evaluating Hindi ASR Systems
This addresses the need for robust multi-accent evaluation in Hindi ASR, which is incremental as it provides a new benchmark and training improvements for a specific domain.
The authors tackled the problem of evaluating Hindi automatic speech recognition (ASR) systems across diverse accents by creating the LAHAJA benchmark with 12.5 hours of audio from 132 speakers, and found that existing models performed poorly, while their model trained on multilingual data with speaker diversity outperformed them significantly.
Hindi, one of the most spoken language of India, exhibits a diverse array of accents due to its usage among individuals from diverse linguistic origins. To enable a robust evaluation of Hindi ASR systems on multiple accents, we create a benchmark, LAHAJA, which contains read and extempore speech on a diverse set of topics and use cases, with a total of 12.5 hours of Hindi audio, sourced from 132 speakers spanning 83 districts of India. We evaluate existing open-source and commercial models on LAHAJA and find their performance to be poor. We then train models using different datasets and find that our model trained on multilingual data with good speaker diversity outperforms existing models by a significant margin. We also present a fine-grained analysis which shows that the performance declines for speakers from North-East and South India, especially with content heavy in named entities and specialized terminology.