Earnings-21

39 hours of 44 English-language earnings calls from 2020 across nine financial sectors, professionally transcribed by Rev.com for benchmarking ASR on named-entity-dense speech.

• Task: asr
• Languages: en
• Hours: 39
• Domain: earnings call
• License: CC BY-SA 4.0
• Homepage: https://github.com/revdotcom/speech-datasets
• Paper: https://arxiv.org/abs/2104.11348

Recommendation

Use as an entity-dense ASR evaluation benchmark for long-form financial/business audio, especially when testing proper-noun and ticker handling. Small at 39h — best as an eval/probe set, not for large-scale training.

Getting the data

Obtain from the dataset homepage.

Suggested processing

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