ISCA Archive Interspeech 2022
ISCA Archive Interspeech 2022

Oriental Language Recognition (OLR) 2021: Summary and Analysis

Binling Wang, Feng Wang, Wenxuan Hu, Qiulin Wang, Jing Li, Dong Wang, Lin Li, Qingyang Hong

The sixth oriental language recognition (OLR) challenge focuses on language identification (LID) and automatic speech recognition (ASR) within multilingual scenarios. OLR 2021 includes four tasks: (1) constrained LID, (2) unconstrained LID, (3) constrained multilingual ASR, (4) unconstrained multilingual ASR. For the LID tasks, Cavg and equal error rate (EER) are considered as primary and secondary metrics, respectively. For the ASR tasks, character error rate (CER) is set as the unique metric. In this challenge, there were 48 participating teams and about half of the teams submitted valid results. Compared to the official baseline systems, the top-ranking teams achieved great progress. Among them, the best submission of constrained LID reduced Cavg from 0.0817 to 0.0025, meanwhile, the unconstrained LID achieved a very low Cavg of 0.0039. As for ASR tasks, the CER was reduced to 13.1% in the constrained ASR and 12.6% in the unconstrained ASR, respectively. This paper describes the above four tasks, database profile, and gives the analysis of results in detail.