Benchmark RadarEarnings25
Earnings25 is a finance-domain benchmark for evaluating automatic speech recognition (ASR) on English-language earnings calls. It includes testset-full (498 hours of S&P 500 earnings calls from Q4 2025) and testset-segmented (46 hours of 290 segments from 2025 U.S. earnings calls), with aligned transcripts and metadata like speaker roles and industry labels, enabling speaker- and industry-aware evaluation beyond WER.
- Released
- 2026-07-26
- Readiness
- Paper only
- Primary field
- Finance & Economics
Why it matters
Existing ASR benchmarks lack domain-specific, large-scale, and realistic finance data, and typically report only aggregate word error rate. Earnings25 provides a reproducible protocol and structured metadata to support more granular evaluation of ASR systems in the finance domain.
Motivation
We introduce Earnings25, a finance-domain benchmark for evaluating automatic speech recognition (ASR) on English-language earnings calls under realistic conditions.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.