Automatisierte Spracherkennung – Audiotranskription
Automated transcription allows spoken content from common audio and video files to be extracted in text form and then further processed and utilized. The Research Data Center supports you by providing advice on free AI transcription tools that can be used, among other things, locally on your own computer.
Open-Source Transcription Tools
noScribe and aTrain are free, open-source apps for Windows, Linux, and Mac designed for automatic audio transcription. Both tools are based on the free AI model whisper support up to 99 languages, and allow users to annotate speakers (e.g., in interview situations or group discussions). Since they run locally on your own computer, they can be used in compliance with the GDPR (no data is exchanged with cloud providers). In addition, both tools offer various export formats (.txt, .html, etc.).
Transcribing Large Datasets
With the tool whisply, which is also based on whisper, you can process and transcribe even large datasets locally in a short amount of time. Instructions for installation and use can be found on whisply’s GitHub page.
With maKI the University of Mannheim also offers an OpenAI-compatible API service that runs entirely on the university’s own infrastructure. Transcriptions of audio and video files can be generated here using NVIDIA’s Parakeet models. Neither requests nor data leave the university.
Services
The FDZ provides support in the following areas, among others:
- General consulting on audio transcription of multimedia content
- Consulting on the use of noScribe and aTrain
- Advice on GDPR considerations regarding audio transcriptions
- Support with further processing of the transcription (e.g., conversion of unstructured data into structured data)
Contact
Forschungsdatenzentrum (FDZ)
Universitätsbibliothek Mannheim
Schloss Schneckenhof West
68161 Mannheim
