FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech
It aims to enable speech technology in more languages and catalyze research in low-resource speech understanding, but is incremental as it builds on existing benchmarks and models.
The paper introduces FLEURS, a benchmark dataset for evaluating few-shot learning in speech across 102 languages, built on FLoRes-101 with about 12 hours of speech per language, and provides baselines using models like mSLAM for tasks such as ASR and translation.
We introduce FLEURS, the Few-shot Learning Evaluation of Universal Representations of Speech benchmark. FLEURS is an n-way parallel speech dataset in 102 languages built on top of the machine translation FLoRes-101 benchmark, with approximately 12 hours of speech supervision per language. FLEURS can be used for a variety of speech tasks, including Automatic Speech Recognition (ASR), Speech Language Identification (Speech LangID), Translation and Retrieval. In this paper, we provide baselines for the tasks based on multilingual pre-trained models like mSLAM. The goal of FLEURS is to enable speech technology in more languages and catalyze research in low-resource speech understanding.