Transcribe interviews automatically

The problem

You've conducted 20 interviews for an evaluation and now you've got 15 hours of audio. Transcribing them manually would take days. You need the text so you can analyse it, but the transcription itself is the bottleneck.

The solution

Use an automatic transcription service like Otter.ai or Trint. Upload your audio files, and they convert speech to text in less time than the recording length. You review and correct any errors, then you've got usable transcripts ready for analysis.

What you get

Text transcripts of each interview, with timestamps and speaker identification. Most services let you play the audio while reading the text, making it easy to check accuracy. You can export as Word, PDF, or plain text.

Before you start

• Interview recordings in common audio/video formats (MP3, MP4, WAV, M4A)
• Consent from interviewees to record (check if consent covers AI transcription)
• Reasonable audio quality - clear speech without heavy background noise

When to use this

• You've got more than 2-3 hours of recordings to transcribe
• Your recordings are reasonable quality with clear speech
• You need text to analyse, quote, or share with colleagues
• Manual transcription would take time you don't have

When not to use this

• Your recordings are poor quality (heavy background noise, multiple people talking over each other, very strong accents)
• You haven't got consent to use third-party transcription services
• The content is extremely sensitive and can't be uploaded to external services
• You only have 1-2 short recordings - might be quicker to just listen and take notes

Steps

1. 1

   Check consent covers transcription

   Review your consent forms. Did you tell participants their recordings would be processed using transcription services? Most services process audio on their servers. If consent doesn't cover this, you may need to go back to participants or use a service with stronger privacy guarantees.

2. 2

   Choose a transcription service

   Otter.ai is good for most use cases and has a generous free tier. Trint is popular in the UK and offers good accuracy. Descript combines transcription with editing. For sensitive content, check where data is stored - some services offer UK/EU data residency.

3. 3

   Upload your recordings

   Create an account and upload your audio files. Most services accept MP3, MP4, WAV, and other common formats. Give files clear names (e.g., 'P01_Interview_2024-01-15') so you can identify them later. Transcription usually takes less time than the recording length.

4. 4

   Review and correct the transcripts

   Automatic transcription is good but not perfect - expect 85-95% accuracy depending on audio quality. Check each transcript against the audio, especially for names, technical terms, and anything you might quote. Most services let you play audio while editing text.

5. 5

   Identify speakers

   Most services attempt speaker identification but often get it wrong. Go through and label speakers correctly (Interviewer, Participant, etc.). This matters if you need to know who said what.

6. 6

   Export in the format you need

   Export as Word doc, PDF, or plain text. Include timestamps if you might need to go back to the audio later. For analysis in an LLM, plain text works best. Keep the originals in the transcription service for reference.

Tools

Resources

Part of this pathway