license: cc-by-4.0
base_model: google/medasr
tags:
- automatic-speech-recognition
- medical
- tactical-combat-casualty-care
- edge-ai
- battlefield-audio
- healthcare
medasr-mil
Fine-tuned version of Google MedASR optimized for Tactical Combat Casualty Care (TCCC) terminology and high-noise battlefield acoustic environments.
Model Description
medasr-mil is a domain-adapted automatic speech recognition (ASR) model designed for combat medic and disaster response scenarios. It improves recognition of critical medical terminology under degraded signal-to-noise conditions including helicopter rotor noise, radio static, and simulated battlefield audio.
This model is intended for research and decision-support system development.
Base model: google/medasr
Intended Use
- Speech-to-text transcription in combat medic scenarios
- High-noise medical environments
- Edge-deployed medical decision support systems
- Research in robust medical ASR
Not intended for:
- Unsupervised clinical decision-making
- Consumer transcription applications
- Diagnostic replacement of human judgment
Training Data
Fine-tuned on 1,260 synthetic and semi-synthetic samples including:
- Tactical Combat Casualty Care terminology
- Simulated helicopter rotor wash
- Radio static overlays
- Gunfire background noise
- SNR degradation (5–20 dB)
- Speed perturbation (0.9–1.1x)
No real patient PHI was used.
Evaluation
Evaluated on 30 held-out tactical audio samples with varied acoustic conditions.
| Metric | Baseline (google/medasr) | medasr-mil |
|---|
| WER | 0.402 ± 0.052 | 0.144 ± 0.043 |
64.2% relative WER reduction under combat acoustic simulation.
Limitations
- Evaluation dataset size (n=30) is limited
- Synthetic noise may not fully represent real battlefield acoustics
- Performance under extreme multilingual scenarios untested
- Not clinically validated
Ethical Considerations
This model is designed strictly for decision-support workflows. It must not replace human clinical judgment.
Deployment in military or emergency contexts should follow appropriate operational and regulatory guidance.
License
Creative Commons Attribution 4.0 International (CC BY 4.0)
