Meddies ASR Benchmark: Vietnamese and English Medical Speech Recognition

A benchmark for automatic speech recognition evaluation, last updated July 2026. It includes four subsets of Vietnamese and English audio, with medical and general-domain slices. The data is packaged as a Parquet file with embedded audio in a datasets.Audio column, created by Meddies.

Use Cases

• Benchmark ASR model performance on Vietnamese general-domain speech based on the deterministic slice from google/fleurs.
• Benchmark ASR model performance on English general-domain speech based on the deterministic slice from google/fleurs.
• Evaluate ASR systems on Vietnamese medical audio sourced from Dr Cao Huu Thinh Official.
• Compare model accuracy across medical and general-domain slices for robustness testing.

Strengths

• Includes four distinct subsets for comparative evaluation.
• Packaged with embedded audio in a datasets.Audio column for direct use.
• Contains a Vietnamese medical audio set from a specific source, Dr Cao Huu Thinh Official.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• References for the Vietnamese medical subset are pending later curation.

Provenance

Source

Subsets derived from google/fleurs and Dr Cao Huu Thinh Official.

Collection Method

Benchmark slices were deterministically selected and packaged.

Freshness

Last updated 2026-07-04 07:56:38; freshness should be verified.