Peer-reviewed research
AISHELL-1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline
Hui Bu et al., arXiv (Oriental COCOSDA 2017), 2017
Read the originalRead on October 17, 2026
- What it is
- The paper that released AISHELL-1, an open Mandarin speech corpus of read speech from many speakers, with a speech recognition baseline. Its results are reported as character error rate, which is the standard metric for Mandarin.
- How much weight it carries
- Peer-reviewed conference paper that is the origin of a widely used open dataset. The baselines date from 2017 and use older acoustic models, so later systems score much lower. Read speech in a studio-like setting is easier than real dictation.
Facts taken from this source
- In the AISHELL-1 baseline study, the best of three 2017 models, LFMMI, reached a character error rate of 7.62 percent on the test set of Mandarin read speech.
“Table 5. Baseline results Model CER of dev CER of test MLLT+SAT 10.43% 12.23% TDNN-HMM 7.23% 8.42% LFMMI 6.44% 7.62%” (Table 5, baseline results)