Task-Lens: Cross-Task Utility Based Speech Dataset Profiling for Low-Resource Indian Languages

A research team led by Swati Sharma has introduced Task-Lens, a novel framework for cross-task utility profiling of speech datasets focused on low-resource Indian languages. Accepted at LREC 2026, this systematic survey addresses the critical challenge of data scarcity in linguistically diverse regions by analyzing 50 existing datasets spanning 26 Indian languages. Unlike traditional single-task catalogues, Task-Lens evaluates datasets across nine downstream speech processing tasks—including automatic speech recognition, speaker identification, and emotion recognition—to uncover hidden potential in existing resources. The framework's core innovation lies in its ability to identify which datasets contain metadata suitable for multiple applications, moving beyond narrow task-specific evaluations to reveal broader utility.

The analysis reveals that many Indian speech datasets contain substantial untapped metadata that could support multiple downstream AI tasks if properly enhanced. Task-Lens provides specific, task-aligned recommendations for dataset improvements and clearly identifies which languages and speech tasks remain critically underserved by current resources. This systematic gap analysis enables researchers to prioritize new dataset creation where it's most needed while maximizing the utility of existing collections. By establishing cross-task linkages and providing a structured methodology for dataset evaluation, the framework represents a significant step toward more efficient resource allocation in multilingual AI development, particularly for regions with rich linguistic diversity but limited digital resources.