WannaGoTech/WannaMedASR
Automatic Speech Recognition • Updated
Datasets:
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The information about the size of the dataset is not coherent.
Error code: UnexpectedError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
model_version_name string | release_date timestamp[s] | base_model string | architecture string | adapter_md5 string | audio_format string | zero_garbage_guaranteed bool | zero_data_leakage_verified bool | comptes_rendus_coverage dict | public_datasets_and_lexicons_integrated dict | training_stages dict | holdout_benchmark_metrics dict |
|---|---|---|---|---|---|---|---|---|---|---|---|
WannaMedASR-v2-3Stage-LoRA-1.7B | 2026-10-29T00:00:00 | Generic-ASR-1.7B-Base | WannaMedASRForConditionalGeneration (28 layers, 196 LoRA matrices = 392 FP32 tensors, r=16, alpha=32) | 02de17701ae6aa8b48def079e5de5de4 | Lossless WAV/FLAC PCM_16 16000 Hz Mono (0% lossy) | true | true | {
"total_reports_trained": 36057,
"historical_reports_purified": 36033,
"surgical_clinical_reports_oct28": 24,
"unique_sentences_extracted": 137267,
"dragon_medical_terms_integrated": 13219,
"doctor_gold_wav_total": 289,
"doctor_gold_wav_trained": 259,
"doctor_gold_wav_holdout_unseen": 30,
"modalities... | {
"Olicorne_UltiMed_Regex_Fixes": {
"total_certified_rules": 41744,
"manual_rules": 1905,
"term_rules": 35876,
"drug_dci_princeps_rules": 3963
},
"AWANNABY_French_Medical_SNOMED_Benchmark": {
"total_rows": 1000,
"holdout_unseen_traps": 30
},
"PxCorpus_Zenodo_6524162_Human_WAV": {
"... | {
"stage_1_global_medical_loss": 0.1074,
"stage_2_surgical_clinic_loss": 0.036,
"stage_3_auto_calibration_loss": 0.0294
} | {
"global_unseen_cer_pct": 0.48,
"global_unseen_wer_pct": 0.82,
"global_unseen_precision_pct": 99.52,
"suite_a_doctor_gold_30_wav_cer_pct": 0.48,
"suite_a_doctor_gold_30_wav_wer_pct": 0.82,
"suite_b_9_modalities_30_cer_pct": 0.41,
"suite_b_9_modalities_30_wer_pct": 0.74,
"suite_c_awannaby_snomed_30_cer_... |
WannaMedASR v2 — French Clinical & Radiological Hold-Out Benchmark (0.48% CER • 99.52% Precision • 0.82% WER)
Official evaluation benchmark and comparative metrics for WannaMedASR v2 (WannaGoTech/WannaMedASR) across French clinical dictation, 9-modality diagnostic imaging (MRI, CT, Ultrasound, Doppler, Mammography, X-Ray, Angiography, RIS), and SNOMED CT pharmacological prescriptions.
- Model Card: WannaGoTech/WannaMedASR
- Interactive Real-Time Space (ZeroGPU A10G): spaces/WannaGoTech/WannaMedASR
- Official Product: https://wannamed.com/asr/ by WannaGoTech (
contact@wannamed.com)
Certified Hold-Out Benchmark Results (Zero Data Leakage — jiwer Exact Levenshtein)
| Model | Clinical Dictation Gold Hold-Out (CER / WER) | 9-Modality Radiology Hold-Out (CER / WER) | Public AWANNABY SNOMED CT + PxCorpus (CER / WER) |
|---|---|---|---|
| WannaMedASR v2 (3-Stage LoRA 1.7B) | 0.48% CER / 0.82% WER (99.52% precision) |
0.41% CER / 0.74% WER (99.59% precision) |
0.56% CER / 0.89% WER (99.44% precision) |
| WannaMedASR v1 (Pass 3) | 1.78% CER / 7.14% WER |
2.26% CER / 9.40% WER |
5.86% CER / 8.22% WER |
Qwen/Qwen3-ASR-1.7B (Base) |
4.14% CER / 15.13% WER |
3.84% CER / 13.10% WER |
15.72% CER / 22.37% WER |
openai/whisper-large-v3-turbo |
14.90% CER / 31.09% WER |
15.86% CER / 43.38% WER |
11.98% CER / 24.34% WER |
openai/whisper-medium |
39.95% CER / 66.39% WER |
40.91% CER / 76.62% WER |
23.94% CER / 42.11% WER |
- Downloads last month
- 5
Models trained or fine-tuned on WannaGoTech/WannaMedASR-French-Clinical-Benchmark