# Earnings-21: entity-dense earnings call ASR benchmark

## Facts

- Provider: Miguel Del Rio, Natalie Delworth, Ryan Westerman et al. (Rev.com)
- License: CC-BY-SA-4.0 (https://github.com/revdotcom/speech-datasets/blob/main/earnings21/README.md)
- Size: 179 files
- Format: JSON Lines (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/Revai/earnings21
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://arxiv.org/abs/2104.11348
- Page: https://scigantic.com/archive/3225f7df-0868-55bc-ad93-6cfa94866758
- Keywords: Speech & Audio, speech recognition, benchmark, JSON Lines, Hugging Face

## About

Earnings-21 is a 39-hour corpus of English earnings calls containing entity-dense speech from nine financial sectors, released by Rev to benchmark automatic speech recognition systems in the wild with special attention to named entity recognition. Calls are long-form, spontaneous and full of company names, people, numbers, dates and financial terms, which makes them a hard test for ASR and for entity-aware word error rate. The release includes reference transcripts in Rev's .nlp token format with entity labels, normalisation files that list acceptable alternative spellings, and per-file WER tags, designed for Rev's fstalign scoring tool. This repository, published by Rev, provides the audio transcoded to 16 kHz in a wav folder, with nlp_references, normalizations and wer_tags folders (44 files each) and a metadata.jsonl index (measured 2026-11-14; the Hub reports about 5.3 GB). Read the wav files with soundfile or torchaudio and the .nlp and JSON references with pandas. The original GitHub repository holds the original audio and full entity tags. Rev releases the Earnings datasets under CC BY-SA 4.0; cite Del Rio et al., Interspeech 2021.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("Revai/earnings21")
print(ds)
```

## Data structure

File types in sampled files: JSON Lines

Example file: `metadata.jsonl`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/3225f7df-0868-55bc-ad93-6cfa94866758
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/3225f7df-0868-55bc-ad93-6cfa94866758
- Over MCP: `get_schema_card` with id `3225f7df-0868-55bc-ad93-6cfa94866758` at https://api.scigantic.com/mcp

## Related topics

- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [JSON Lines datasets](https://scigantic.com/topics/json-lines)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
- [Speech recognition datasets](https://scigantic.com/topics/speech-recognition)
