The Veranda Interview: Clipping, Hum and a Highway
Priya is recording a six-minute oral-history interview with a 70-year-old folk singer on his veranda, 40 m from a state highway, for a college radio documentary. Her kit is one shotgun mic on a boom, one lavalier, a two-channel recorder, closed-back headphones and a wind muff. On the first take, truck whooshes dominate, the singer's voice peaks at +2 dBFS twice, a low hum appears whenever the recorder is plugged into mains power, and hand-claps on the veranda's tin roof ring for about 1.5 seconds. The singer leaves town on Thursday evening; today is Wednesday. Priya must deliver a 48 kHz/24-bit WAV and a script-ready transcript by Friday, and she has no second recordist.
1. Diagnose the clipping and the hum, and state the recording settings and power arrangement you would use for the retake.
Read the worked answer
Clipping at +2 dBFS is unrecoverable. Once the waveform hits the ceiling, the tops of the loudest syllables are flattened into square edges, and that harmonic distortion cannot be undone in post - a limiter can hide it, not repair it. Those two moments must be re-recorded.
Settings for the retake:
- Recorder at 48 kHz / 24-bit. Bit depth is the headroom: 24-bit gives a very low noise floor, so recording quietly costs nothing.
- Aim for speech peaks around −12 dBFS and average speech around −18 to −20 dBFS. That leaves ~12 dB of safety for a sudden laugh or a raised voice.
- Use the limiter as a backstop, not as a level strategy, and set it below 0 dBFS (e.g. −3 dBFS) so it almost never engages.
- Set gain by having the singer speak his loudest passage before rolling; lock gain and do not ride it mid-answer.
- Monitor on closed-back headphones throughout - the clipping happened because nobody was watching the meters.
The hum is a ground loop. It appears only on mains power, which points to two earth paths (recorder and building wiring). Run the recorder on batteries for the whole session, and test by recording 30 seconds on battery and 30 on mains and comparing. If mains power is unavoidable, use a single earthed supply and avoid touching the mic body while recording. A 50 Hz notch filter exists, but it eats the singer's low register - fix the cause, not the symptom.
2. Traffic noise and a ringing tin roof are wrecking an intimate interview. What do you change in the setup, given you cannot move the veranda?
Read the worked answer
Traffic. At 40 m from a highway, the noise is continuous and broadband, so software noise reduction will leave artefacts and hollow speech if relied on. The gains must come from placement and timing:
- Move the singer to the far end of the veranda, ideally with a wall or building between him and the road, and orient the mic so the road is at its off-axis null - point the mic's back or side at the traffic.
- Work closer: 20–25 cm mic-to-mouth reduces the ratio of road noise to voice substantially.
- Schedule around traffic peaks - record between the heavy-vehicle waves, and take one full answer at a time rather than pushing through.
- Keep the wind muff off indoors; foam and fur dull the high frequencies that give speech clarity.
Ringing roof. A shotgun indoors is the wrong tool - its narrow pattern picks up the first reflections it hears and off-axis colouration makes voices sound hollow. Switch to the lavalier, clipped 15–20 cm below the chin, or a cardioid on the boom just above and slightly to the side of the mouth. Sit the singer with his back to a soft surface (a curtain, a bedding roll, a quilt) so the immediate reflection is absorbed. A handheld moving blanket behind the mic position works for a tin-roof veranda.
Finally, record 60 seconds of room tone at each setup - the ambient 'silence' - so the editor can build a noise bed under the cuts and keep the traffic sounding continuous rather than pumping in and out.
3. With the singer leaving Thursday evening and a Friday deadline, lay out the session plan and the file and transcript workflow.
Read the worked answer
Session plan for today (Wednesday):
- Do not spend the visit re-running what already exists. Make a short list of the damaged sections - the two clipped answers plus anything drowned by traffic - and re-record those answers in full, not patched sentences.
- Correct the setup before rolling: battery power, lavalier or cardioid, softer backdrop, road off-axis. Record a 30-second test and listen to it on headphones before the singer is asked to perform.
- Re-take the room tone, and ask for a second, slower pass of the name of the village, the song titles and the years - those are the details a transcript must spell correctly.
- Before leaving, play back the take with the singer present so he can correct a date or a name while he is still in the room. Tomorrow he is gone.
File and safety workflow:
- Folder: project name / date / subject; file names that carry the subject, date and take number.
- Two copies before the card leaves the house - internal drive plus an external drive; clone, then verify by playback, and only then clear the card.
- Export a rough 48 kHz/24-bit WAV of the whole session for transcription; keep the raw recorder files untouched as masters.
Transcript: mark timecodes every 30–60 seconds, note unclear words rather than guessing, and flag anything the singer corrected during playback. Deliver the WAV at 48 kHz/24-bit with the room tone as a separate file, plus a consent note for the oral-history archive.