LibriPhrase, introduced in our paper “Learning Audio-Text Agreement for Open-vocabulary Keyword Spotting”, has gained widespread recognition as a trusted benchmark for user-defined keyword spotting (UDKWS) and open-vocabulary keyword spotting (OV-KWS) tasks.

Our research proposed a novel approach that focuses on the agreement of phonetic information between spoken speech and text sequence for keyword spotting system. It’s truly inspiring to see how our work has motivated numerous researchers to build upon our framework. We are happy that our research has positive impact on the keyword spotting field.

Moreover, we’ve recently released the official LibriPhrase benchmark dataset, as featured in our paper, available for wider use. This dataset provides an opportunity for you to evaluate your systems alongside state-of-the-art solutions.

If you’re interested in exploring this dataset, please follow this link to access the LibriPhrase benchmark set:

LibriPhrase Github