Google MedASR

MedASR is a speech-to-text model based on the Conformer architecture, pre-trained for medical dictation and transcription. Developed by Google, it contains 105 million parameters and was trained on approximately 5,000 hours of de-identified physician dictations spanning multiple specialties including radiology, internal medicine, and family medicine. It accepts mono-channel audio (16kHz, int16 waveform) and generates text-only transcriptions. MedASR is recommended for dictation tasks involving specialized medical terminologies and can be fine-tuned for accents, acoustic environments, vocabulary expansion, and formatting. It integrates with generative models like MedGemma for summarization and question answering.