Today I learned that using whisper.cpp and your CPU, you can transcribe german audio via -l de, using the standard model weights, available via huggingface. For location in the manual documentation see.

Suppose your repo is at /path/to/repo/whsiper.cpp, then full command to execute in the root folder of whisper.cpp repo is: ./main -m /path/to/repo/whisper.cpp/models/ggml-medium.bin -l de -t 12 -pc -oj -f samples/your-audio.wav