Transcription
There are different approaches to transcription, and although I work in a multimodal framework, I will here describe how to facilitate transcription of recorded voice only. All software used is free and/or already included in your computer and the transcription process is offline.
Computers come with tools for speech-to-text. These are mostly used for voice control, but are essentially AI:s for speech recognition. My idea was to use this function to reduce the time I spend on manually transcribing voice recordings. Using the built-in tools enable me to work offline and there is no need to use third party transcription services. I need to ensure the integrity of my research participants and my data and an offline solution was the safest option I could think of.
The AI generated text is incomplete and incorrect but it reduces the time I spend on transcription substantially. Please note that the first automatically generated text is only raw material for a proper transcription. It cannot do the job for you as if you had used a paid transcription service.
I used a Mac computer (ios) for the process I describe below as it can be used for multiple languages. There is a good AI for PCs (windows) but mine only recognises English, and in this case the voice recording was in another language.
How to automatically generate text from a voice recording
This is how to do it on a Mac (but the process is essentially the same on a pc). You will need to download Audacity and Soundflower both are free programs and available for both mac ios and windows.
Please note that no links are added but only the adresses as text*. You can cut and paste the link adress into any browser you are using.
1) Install programs
- Audacity can be found at audacityteam.org/download/
- Soundflower can be found at soundflower.en.softonic.com/mac. (If you run into problems with the installation you can read more at techwiser.com/record-internal-audio-on-mac/)
- If you have persistent problems installing soundflower there is a program called the unarchiver (theunarchiver.com) that did the trick for me. I unpacked the soundflower installation pack first and then went back and followed the installation procedure outlined in thetechwiser.com/record-internal-audio-on-mac/
2) If necessary reformat** the sound file to mp3
3) Change settings for dictation so that output and input for sound is soundflower 2ch
4) Transcribe
- Open mp3 file in audacity
- Open notepad (notes worked better than pages for some reason)
- Play sound from audacity and start dictation with double clicking fn
The dictation seems to be able to cope with about 40 seconds of voice before it gets lost so I had to keep restarting every 30 seconds keeping track of where I was in Audacity. I also slowed down the speed to about 90% which seemed to make the dictation work better (but I did no comprehensive trial here).
This solution allows me to work offline keeping the integrity of my data and participants intact. I don’t have consent for any third party tools such as google or other online solutions.
*No active links are included in this blog post. Please refer to the EU 2001 Information Society Directive, and 2019 Directive on Copyright in the Digital Single Market.
**One way to reformat (convert) a sound file to mp3 is by using vlc player. This ensures that you are working offline maintaining data integrity.
- Open the file in vlc on a pc
- Install vlc if necessary. Vlc player can be found at videolan.org/vlc/
- Open stream/convert
- Save as mp3 but ALSO manually rename the file as mp3 (edit the file extension)
multimodal study 