ClinicalWhisper: A privacy-first local pipeline for multimodal clinical speech analysis
ClinicalWhisper is an MIT-licensed, local-first research pipeline, built to study anhedonia, that converts sensitive audio into speaker-attributed transcripts with identifiers masked. It runs as a desktop app on Apple Silicon Macs, and recordings can be batch-processed from the command line. Researchers who only need local transcription can install the core app alone, built on MOSS-Transcribe-Diarize; optional add-ons give exploratory clinical scores that have not been externally validated. Each recording's full results are saved as a JSON file, and a batch also produces a summary CSV with one row per recording, so measures from many interviews can go straight into statistical analysis. This software paper describes ClinicalWhisper 5.3 and is to be submitted to the Journal of Open Source Software. Source code: https://github.com/czhou732/Clinical-Whisper-Pipeline
Authors
- Chengdong Zhou
- Jinghao Yu (ORCID: https://orcid.org/0009-0008-4054-6020)
Institutions
- University of Southern California (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24286717
- Primary Topic
- Speech Recognition and Synthesis
- Type
- preprint