Project-W is an open-source, self-hostable platform designed to transcribe audio files (speech-to-text) with a strong emphasis on data privacy and ease of use. Built to handle sensitive content such as research interviews or medical recordings, Project-W enables researchers and non-technical users to process audio securely within their organization's infrastructure - without relying on third-party services or needing to perform complex installations or setups themselves.At its core, Project-W leverages OpenAI's Whisper model for transcription, while also offering advanced features such as speaker diarization (automatically distinguishing between speakers), timestamp alignment (linking text segments to their corresponding moments in the recording), translation, and support for multiple output formats including subtitle files for video use.All of these capabilities are accessible through a streamlined, easy-to-use web interface that requires no setup or technical expertise from the user. In addition, Project-W provides a well-documented REST API for integration into automated workflows or use by alternative clients.To simplify hosting and integration into institutional environments, Project-W provides full support for both LDAP and OIDC authentication in addition to local accounts, with the flexibility to combine multiple authentication providers. Administrators can also configure data retention policies (automatic data and user deletion), terms of services, and certain interface elements for full compliance with legal requirements and their local standards.Project-W is currently alpha software and can be tested out on the following public instance: https://project-w.urz.uni-heidelberg.de
Partanen et al. (Thu,) studied this question.