Sanskrit ASR Benchmark Dataset: Noisy Speech Recognition

About Dataset

This is a Sanskrit ASR benchmark dataset specifically designed to evaluate and improve Automatic Speech Recognition (ASR) systems in noisy and challenging scenarios, particularly in the general domain. The dataset comprises 1684 hours of labeled speech data across 12 Indian languages, with a focus on Sanskrit. This dataset variant, known as "Kathbath-Sanskrit-Noisy-Test-Known," provides researchers and developers with a valuable resource for building robust ASR models capable of handling real-world noisy conditions. Submitted by Tahir Javed, it supports advancements in speech recognition technologies for regional languages.