eka-hard / README.md
ronanarraig's picture
Upload README.md with huggingface_hub
018d085 verified
metadata
license: mit
task_categories:
  - automatic-speech-recognition
language:
  - en
tags:
  - medical
  - asr
  - entity-cer
  - benchmark
size_categories:
  - n<1K

EKA Hard — Medical ASR Benchmark

Entity-aware medical ASR benchmark — 50 hard rows from Indian-accented clinical speech.

Prepared by Trelis Research. Watch more on Youtube or inquire about our custom voice AI (ASR/TTS) services here.

Source

Derived from ekacare/eka-medical-asr-evaluation-dataset (3,619 EN rows, MIT license). Real clinical speech from 57 speakers across 4 Indian medical colleges, 16kHz mono.

Preparation

  1. Filter: audio ≥ 2s, text ≥ 20 chars
  2. Gemini Flash entity tagging (6 medical categories)
  3. Keep rows with ≥ 1 entity
  4. 3-model difficulty filter (whisper-large-v3, canary-1b-v2, Voxtral-Mini) with whisper-english normalization
  5. Top-50 by median entity CER

Entity categories

  • drug — drug or medication names (brand or INN)
  • condition — diagnoses, diseases, syndromes, disorders
  • procedure — surgical, diagnostic, or therapeutic procedures
  • anatomy — anatomical structures, organs, body regions
  • biomarker — lab tests, biomarkers, genes, proteins, molecular markers
  • organisation — hospitals, regulatory bodies, pharmaceutical companies

Columns

  • audio — 16kHz WAV
  • text — ground truth transcript (human-annotated)
  • entities — JSON array of tagged medical entities with text, category, char_start, char_end
  • difficulty_rank — 1 = hardest
  • median_entity_cer — median entity CER across 3 difficulty-filter models

Leaderboard (16 models, sorted by Entity CER)

# Model WER CER Entity CER Results
1 gemini-2.5-pro 0.150 0.078 0.210 results
2 scribe-v2 0.273 0.154 0.279 results
3 parakeet-tdt-0.6b-v3 0.376 0.206 0.309 results
4 ursa-2-enhanced 0.341 0.237 0.314 results
5 universal-3-pro 0.434 0.337 0.353 results
6 nova-3 0.449 0.291 0.387 results
7 canary-1b-v2 0.398 0.224 0.392 results
8 whisper-large-v3-turbo 0.351 0.216 0.394 results
9 whisper-v3 (fireworks) 0.439 0.268 0.414 results
10 Voxtral-Mini-3B-2507 0.439 0.295 0.426 results
11 MultiMed-ST (whisper-small-en) 0.491 0.351 0.450 results
12 whisper-base 1.268 0.789 0.472 results
13 medasr 0.627 0.453 0.478 results
14 whisper-tiny 1.398 0.780 0.572 results
15 whisper-large-v3 1.060 0.569 0.757 results
16 whisper-small 5.201 2.782 0.946 results

Evaluated with Trelis Studio, whisper-english normalization.