Preprint
Earnings-22: A Practical Benchmark for Accents in the Wild
Miguel Del Rio et al., arXiv (Computation and Language), submitted to Interspeech 2022, 2022
Read the originalRead on October 17, 2026
- What it is
- The paper that released Earnings-22, a free corpus of English earnings calls from companies in many countries. It was built to test speech recognition on accented, real-world audio, and it compares four commercial systems on it.
- How much weight it carries
- A preprint submitted to Interspeech 2022. The authors built the corpus and also compare four commercial systems on it, so the comparison should be checked independently. The corpus is free to use, and its speech is spontaneous English from financial calls, one genre among many.
Facts taken from this source
- Earnings-22 is a corpus of 125 files totaling 119 hours of English earnings calls from companies around the world.
“we present Earnings-22, a 125 file, 119 hour corpus of English-language earnings calls gathered from global companies” (Abstract)
- The Earnings-22 paper says modern speech recognition systems have reached superhuman word error rates on many common corpora, while lacking adequate performance on speech in the wild.
“Modern automatic speech recognition (ASR) systems have achieved superhuman Word Error Rate (WER) on many common corpora despite lacking adequate performance on speech in the wild.” (Abstract)