NVIDIA Clinical ASR — настройка медицинского распознавания речи

★ 7.6 · content

digital-health-clinical-asr-setup is a Claude Code skill that bootstraps Stage 1 of NVIDIA's Clinical ASR Flywheel: it verifies the NVIDIA_API_KEY, installs Python dependencies (Python 3.10+), and runs a smoke test that round-trips a clinical sentence through Magpie TTS and Parakeet/Nemotron ASR over NVCF. Before any text or audio is sent, the skill surfaces a mandatory disclosure covering two outbound data flows — NVIDIA NVCF (synthetic sentences and WAV files) and Merriam-Webster (individual clinical terms for IPA tagging). The entire flywheel targets KER (keyword error rate) on clinical entities such as drugs, procedures, anatomy, and diagnoses, because WER averages obscure clinically significant errors. Once the smoke test passes, the user is cleared to advance to digital-health-clinical-asr-build. The pipeline operates on synthetic data only; real patient transcripts or PHI must never be fed into any stage.