TranscriboZH – Open Source Audio Transcription
In collaboration with the Public Prosecutor’s Office of the Canton of Zurich, the Zurich Cantonal Police, the Zurich City Police and the Winterthur City Police, a team from the Statistical Office of the Canton of Zurich has developed a prototype app that automatically transcribes confidential audio and video files. Also in Swiss German.
The app is open-source and based on the Whisper v3 Large model, which enables transcriptions up to 15 times faster than in real time – without license or usage costs. The app offers a wide range of functions such as automatic speaker recognition, multi-file upload, predefined vocabulary and various export options. Transcripts can be edited directly in the application and linked synchronously with the source file.
Hardware requirements: recommend using a CUDA-compatible graphics card with at least 8GB VRAM, as transcription on a CPU is extremely slow.
Source: https://github.com/machinelearningZH/audio-transcription.