Recording an interview cleanly — settings for interviews and conversations

An interview does not come round twice. Once it starts there is no room to fiddle with settings, so this is what to decide before you sit down.

The recording screen, with the record button and elapsed time
The recording screen

Decide before you start

Where to put the device

Put it at roughly the height of the speakers' mouths, slightly closer to the other person than to the middle. Your own voice arrives loud because you are close to it, so leaning the placement towards them balances the two. In a café, take the wall side rather than the aisle: every person walking past brings footsteps and conversation with them.

Format and bit depth

Choose WAV at 24-bit. Interviews swing hard between someone leaning in and getting loud and someone thinking aloud and going quiet, and 16-bit tends to crush the quiet side. With 24-bit you can lift it afterwards without the noise coming up with it.

Whether to add a microphone

An external microphone reliably sounds better, but it can also make people self-conscious. A device lying on the table often gets a more natural conversation, so judge it by the kind of interview you are doing. Avoid Bluetooth microphones for long sessions: if the connection drops, so does the recording.

The 30 seconds before you start

  1. Start recording and check that the input level moves
  2. Say out loud that you are recording — keep that sentence in the recording
  3. Switch on airplane mode. An incoming call stops the recording
  4. Check the free space. Two hours at 24-bit runs to several GB

Consent that is captured in the audio can be checked later even without anything on paper. Saying the date, the place and the person's name at the top also makes the file easy to find once you have a lot of them.

While it runs

Recording continues with the screen off. When something worth using comes up, it is faster to look at the clock and write down the time than to take notes. Transcripts come out with timestamps, so that number takes you straight there.

Afterwards

Run the whole thing through a light model first to get the shape of the conversation. Once you know which parts you will quote, redo just those with a heavier model. Putting the entire recording through a heavy model means waiting a long time to improve accuracy on material you will never use.

Names break. Write down the company, product and personal names before you transcribe, and check them against the output afterwards.

Turning it into copy

Cleaning up the text removes hesitations and repetitions and gets you close to something readable. But the way someone speaks is information. Smoothing all of it out erases the person, so decide per quote whether to keep the original phrasing, based on what the piece is.

Handing it over and keeping it

If you need to give the audio to an editor or a client who is in the same room, AirDrop is the quickest route and never touches a cloud service. Transcripts can go as a text file. Interview audio usually contains personal information, so agree in advance who receives it and how long it is kept.

Where it usually goes wrong

Only the other person sounds quiet

The device was sitting closer to you. Move it towards them next time. For a recording you already have, a heavier model can sometimes still pick the words out.

The music in the venue drowns everything

Music and speech occupy the same frequencies, so separating them afterwards is close to impossible. This one is decided when you choose the table. Do not sit under a speaker.

I am not sure a long interview will fit

Turn on splitting at a fixed interval. It makes the space easier to predict, and if something goes wrong you lose one file rather than everything.

Related

Get the app

Android Get it on Google Play iOS Coming soon to the App Store