---
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](https://youtube.com/@TrelisResearch) or inquire about our custom voice AI (ASR/TTS) services [here](https://trelis.com/voice-ai-services).


## Source

Derived from [ekacare/eka-medical-asr-evaluation-dataset](https://huggingface.co/datasets/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](https://huggingface.co/datasets/Trelis/eval-gemini-2.5-pro-eka-hard-20260408-1922) |
| 2 | scribe-v2 | 0.273 | 0.154 | 0.279 | [results](https://huggingface.co/datasets/Trelis/eval-scribe-v2-eka-hard-20260408-1924) |
| 3 | parakeet-tdt-0.6b-v3 | 0.376 | 0.206 | 0.309 | [results](https://huggingface.co/datasets/Trelis/eval-parakeet-tdt-0.6b-v3-eka-hard-20260408-1920) |
| 4 | ursa-2-enhanced | 0.341 | 0.237 | 0.314 | [results](https://huggingface.co/datasets/Trelis/eval-ursa-2-enhanced-eka-hard-20260408-1924) |
| 5 | universal-3-pro | 0.434 | 0.337 | 0.353 | [results](https://huggingface.co/datasets/Trelis/eval-universal-3-pro-eka-hard-20260408-1924) |
| 6 | nova-3 | 0.449 | 0.291 | 0.387 | [results](https://huggingface.co/datasets/Trelis/eval-nova-3-eka-hard-20260408-1924) |
| 7 | canary-1b-v2 | 0.398 | 0.224 | 0.392 | [results](https://huggingface.co/datasets/Trelis/eval-canary-1b-v2-eka-hard-20260408-1921) |
| 8 | whisper-large-v3-turbo | 0.351 | 0.216 | 0.394 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-large-v3-turbo-eka-hard-20260408-1921) |
| 9 | whisper-v3 (fireworks) | 0.439 | 0.268 | 0.414 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-v3-eka-hard-20260408-1928) |
| 10 | Voxtral-Mini-3B-2507 | 0.439 | 0.295 | 0.426 | [results](https://huggingface.co/datasets/Trelis/eval-Voxtral-Mini-3B-2507-eka-hard-20260408-1920) |
| 11 | MultiMed-ST (whisper-small-en) | 0.491 | 0.351 | 0.450 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-small-english-eka-hard-20260408-1925) |
| 12 | whisper-base | 1.268 | 0.789 | 0.472 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-base-eka-hard-20260408-1921) |
| 13 | medasr | 0.627 | 0.453 | 0.478 | [results](https://huggingface.co/datasets/Trelis/eval-medasr-eka-hard-20260409-1126) |
| 14 | whisper-tiny | 1.398 | 0.780 | 0.572 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-tiny-eka-hard-20260408-1921) |
| 15 | whisper-large-v3 | 1.060 | 0.569 | 0.757 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-large-v3-eka-hard-20260408-1922) |
| 16 | whisper-small | 5.201 | 2.782 | 0.946 | [results](https://huggingface.co/datasets/Trelis/eval-whisper-small-eka-hard-20260408-1924) |

Evaluated with [Trelis Studio](https://studio.trelis.com), whisper-english normalization.