The interviews were conducted in the second week of January. There are a number of options for obtaining an accurate transcription. The traditional method is to transcribe the recordings manually. This is possible, but generally takes 6 to 8 hours for each hour of recording. It is easier to run the audio through a modern speech recognition programme and then check and correct the recognition results if necessary. This also allows you to indicate who spoke when.

There are two options for automatic speech recognition: in the cloud or on your own computer. 

Cloud

A nice and simple way to transcribe speech is via the CLARIN-LMU portal in Munich. However, you must first log in with your academic account. SURF also has a cloud service, but it will not be available until spring 2026.

Your own computer

If you don't have a lot to recognise, it's probably just as easy to do it on your own computer. To do so, however, you must first download and install the software. There are now many Whisper-based speech recognisers, so which one should you use?

Python

People with a good knowledge of Python can use the original Whisper or WhisperX. 

noScribe

Others are best off using noScribe. This is an open source Whisper speech recogniser that offers excellent results (for both recognition and speaker diarisation). The disadvantage is that the output format is limited (only text or vtt for subtitling).

MacWhisper

If you are serious about speech recognition, I would recommend MacWhisper. It costs around 64 euro but can only be used on an Apple computer. 

aTrain

Windows users can use aTrain (free of charge). Speech can also be recognised using the open source packages GoldWave and Subtitle Edit.

Send it to me

A final option is to send the audio to me and I will do the recognition.