ASR Leaderboard Longform: Three Benchmark Test Sets for Speech Recognition

ASR Leaderboard Longform provides three standardized benchmark test sets—Earnings-21, Earnings-22, and TED-LIUM—for evaluating longform automatic speech recognition models. The dataset is hosted by hf-audio on Hugging Face and was last updated on June 11, 2026. It is formatted as Parquet files for efficient loading via the Hugging Face datasets library.

Use Cases

• Benchmark ASR model performance on longform audio based on the provided Earnings-21 and Earnings-22 test sets.
• Evaluate model robustness on real-world, extended speech segments based on the dataset's stated design.
• Compare model accuracy across different speech domains using the distinct TED-LIUM and Earnings test sets.

Strengths

• Includes three distinct longform test sets: Earnings-21, Earnings-22, and TED-LIUM.
• Data is provided in a standardized Parquet format for safe, efficient loading.
• Designed for evaluating models under real-world conditions like extended audio segments.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-06-11).

Provenance

Source

hf-audio on Hugging Face

Collection Method

Likely compiled from existing benchmark sources (Earnings calls, TED talks) for leaderboard evaluation.

Freshness

Last updated 2026-06-11 09:24:26