Why customer interview recordings pick up so much noise
A customer interview is usually recorded in the worst possible acoustic setup: a laptop mic across a kitchen table, a phone on speaker, or a Zoom call where one person is in a hotel and another is next to a window. You get HVAC rumble, keyboard clicks, mouse clicks, chair squeaks, traffic, fan noise, notification dings, and room echo. The result can hurt transcription accuracy and make you sound amateur.
You usually do not need to re-record. You need a noise-removal workflow that separates constant noise from one-off noises, then processes each in the least destructive order. Below is a practical sequence for remote meetings, webinars, interviews, podcasts, phone recordings, gaming clips, and voice memos.
Step 1: Identify the noise before you remove it
Open the file in any editor that shows a waveform. Play the first five seconds before anyone speaks. That is your room tone. Is the noise steady, like a hum or fan? Or intermittent, like keyboard clicks and dog barks? This matters because a noise gate helps with clicks, while spectral noise reduction helps with hum and hiss. A high-pass filter helps with rumble. De-reverb helps with echo. Using the wrong tool is how recordings end up sounding underwater.
Listen with headphones, not laptop speakers; laptop speakers hide low rumble and harsh denoise artifacts. Note the noise type in the filename or a notes app.
Step 2: Work on a copy and clean up the obvious stuff first
Never process your only original. Duplicate the file and keep the original untouched. If your recorder offers WAV, use it; MP3 and AAC discard data that noise removal needs. Before any AI or denoise tool, do manual repairs:
- Cut or mute loud one-off noises between words, like a cough, door slam, or notification.
- Delete long silences, but keep a half-second of room tone on each side so edits sound natural.
- Remove overlapping filler if it does not change meaning, but do not cut a customer off mid-thought.
- Use short crossfades (5-15 ms) at every edit point to avoid clicks.
This step alone can make a noisy interview sound dramatically better because denoise tools have less garbage to chew on.
Step 3: Use a noise gate or expander for intermittent clicks
A gate mutes the track when the signal drops below a threshold. It is useful for keyboard taps, mouse clicks, and paper rustling. Set the threshold just above the room noise but below quiet speech. In a typical interview, start around -40 dB to -35 dB. Set attack to 1-5 ms, hold to 100-200 ms, and release to 100-250 ms. If the gate cuts off word tails or makes breaths disappear, lower the threshold or increase release. A gate is not a magic noise remover for constant hum; it will not clean the noise during speech.
Step 4: Apply gentle noise reduction with a noise profile
For steady hiss, fan, or air conditioning, use a noise reduction tool that learns a noise profile. In Audacity, select one to two seconds of room tone, open Effect > Noise Removal and Repair > Noise Reduction, and click Get Noise Profile. Then select the whole interview. Start with 6-12 dB of reduction, sensitivity 6, and frequency smoothing 3. Apply it once. If the noise remains, apply a second pass with 3-6 dB instead of one heavy pass at 20 dB. Heavy settings create metallic, robotic, or lispy speech.
If the noise is a hum, set a high-pass filter at 80-100 Hz for most voices. If you hear a 50 Hz or 60 Hz electrical hum, use a narrow notch filter at that frequency and its harmonics (100/120 Hz, 150/180 Hz). Do not high-pass above 120 Hz unless the speaker has a very thin voice.
Step 5: Try an AI or online cleanup for remote meeting audio
AI noise removal has gotten very good at handling Zoom calls, webinars, and interview recordings where the noise is complex: traffic, cafe chatter, fan, and room echo in one file. A free tool like ClearVoiceHQ (clearvoicehq.com) lets you upload a recording and get a cleaner version without installing a full editor. It is a practical first pass when you need a quick result, especially for a one-off customer interview or voice memo. Upload the file, run the cleanup, download the result, and then compare it with the original at the same volume. If it sounds over-processed, use a lighter setting or fall back to manual noise reduction.
ClearVoice can also help when you have several recordings from different remote guests and want a consistent baseline before editing. Treat it as one stage in the chain, not the only stage. You still need manual edits for clicks, long pauses, and level balancing.
Step 6: Fix specific recording types
Zoom, Google Meet, and Microsoft Teams
If you recorded locally, you often have separate tracks. Process each speaker separately. That lets you use a stronger gate on the person with the mechanical keyboard and a lighter denoise on the person with fan noise. If you only have the cloud recording, use denoise first, then manual edits, then EQ.
Phone voice memos and customer support calls
Phone audio is compressed and band-limited. It already lacks low bass and high sparkle. Use a high-pass at 100 Hz, gentle denoise, and a de-esser if 's' sounds harsh. Avoid heavy noise reduction because phone speech breaks down quickly.
Podcasts and interview clips for marketing
Process the voice before music or ads. Target around -16 LUFS for mono podcast audio or -14 LUFS for stereo video. Use a limiter only after cleanup. Export a WAV master and a compressed MP3 or AAC for sharing.
Gaming clips and screen recordings
Keyboard and mouse clicks are transient. Use a gate or expander before denoise. If game audio is on the same track as voice, you cannot fully remove background noise without removing the game. Separate tracks next time; for now, use a gate and accept some compromise.
Step 7: Check the result like an editor
Listen at low volume and then at normal volume. At low volume, artifacts and pumping are easier to hear. Solo the track and ask: Do consonants sound natural? Are breaths gone? Does the noise floor breathe in and out? Does the customer still sound like themselves? Run a transcription pass. If the transcript is cleaner, your noise removal helped. If names and numbers got worse, you over-processed.
A good A/B test is to export two versions: original and cleaned. Play them back-to-back at matched loudness. The cleaned version should be easier to understand, not just quieter. If the noise is less annoying but the voice sounds thin, back off the denoise and high-pass filter.
Fast workflow for a single customer interview
- Duplicate the original file.
- Remove obvious clicks, coughs, and long gaps manually.
- Apply a high-pass filter at 80-100 Hz.
- Use a noise gate at -40 to -35 dB for intermittent clicks.
- Use noise reduction with a learned profile at 6-12 dB, or run a free online tool like ClearVoice for complex noise.
- Notch out 50/60 Hz hum if needed.
- Normalize to -3 dB peak, then limit to -1 dB true peak for video or -16 LUFS for podcast.
- Export and listen on phone speakers and headphones.
Noise removal is a tradeoff. The goal is not silence. The goal is a clean, understandable interview where the customer's voice remains intact. Keep the original, make small moves, and compare. With a sensible chain, you can rescue a noisy customer interview recording without asking anyone to record it again.