Earnings-22

119 h benchmark of real-world English corporate earnings calls featuring diverse global accents across many countries.

• Task: asr
• Languages: en
• Hours: 119
• Domain: earnings calls
• License: CC BY-SA 4.0
• Homepage: https://github.com/revdotcom/speech-datasets/tree/main/earnings22
• Paper: https://arxiv.org/abs/2203.15591

Recommendation

Use as an evaluation benchmark for accented, real-world long-form English ASR rather than a large training set — ideal to stress-test robustness across non-native and regional accents. Small (~125 files) and benchmark-oriented, with rich per-file metadata.

Getting the data

Obtain from the dataset homepage.

Released by Rev.com; speakers span 7 language regions / 27 countries.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/asr-training-data.yaml — run it with vkit docker run.