Gaps, Clicks and Stutters: Tracking Down Audio Dropouts

What a dropout sounds like

"Dropout" covers several symptoms, and each hints at a different part of the chain. Listen on headphones and note which ones you hear.

• Silent gaps. The audio cuts to nothing, or to a different background sound, for a fraction of a second or longer.
• Clicks and pops. Sharp ticks where the waveform jumps abruptly, sometimes at regular intervals.
• Stutters. A tiny fragment repeats, as if the recording skipped back for an instant.
• Robotic or metallic patches. Words become warbly or synthetic, typical of internet calls when data goes missing.
• Bursts of noise or hiss. A crackle or rush of static, common with analog wireless systems and damaged cables.
• Level dips. The voice briefly drops in volume, then returns, often from a radio signal fading or a body blocking a transmitter.

The usual causes

Wireless interference

Distance, walls and bodies between transmitter and receiver, crowded radio bands shared with Wi-Fi and Bluetooth, low batteries

Buffer underruns

The computer cannot deliver audio to or from the interface in time because of CPU load, small buffer settings or USB power saving

Storage problems

A memory card too slow for the recording format, nearly full, worn out or corrupted

Cables and connectors

A loose plug, worn cable or intermittent phantom power that crackles when touched or moved

Network packet loss

Data lost or delayed on an internet call; the call software hides gaps by stretching or synthesising sound

Sample rate and clock mismatches

Two devices or apps running at different rates, or separate devices with unsynchronised clocks, forcing constant resampling

Wireless systems behave differently by type. Analog systems tend to produce bursts of noise as the signal weakens, while digital systems tend to stay clean until the signal fails and then mute or glitch. Bluetooth headsets add another failure: switching into a lower-quality call mode, or dropping the connection, partway through a recording.

Diagnose from the pattern

The timing of the dropouts is the most useful clue. Open the file in an editor such as Audacity, zoom into a few dropouts, and compare what you see with what you hear.

• Regular clicks at fixed intervals point to a clock or sample rate mismatch, or to a buffer that is consistently too small.
• Random gaps that coincide with the speaker moving point to wireless range, body blocking or a cable being tugged.
• Problems that start partway through and get worse suggest a battery running down, a card filling up or a computer heating up.
• Robotic patches only in remote voices, never in the local one, point to the network rather than your equipment.
• A flat line at exactly zero means samples were lost or replaced with silence by the recording device; a gap that still contains room tone means the microphone or transmitter muted.
• A file shorter than the real event means the recorder dropped data rather than recording silence.

Steps to confirm the cause

1. Check for backups first: camera audio, a second recorder, a phone recording, or each participant's local recording of a call.
2. Note the time of every dropout in the first ten minutes and look for a regular interval.
3. Check the recording settings: sample rate in the interface, the operating system and the recording app should match.
4. Review the conditions: battery levels, distance from receiver, card type and free space, other apps running.
5. Reproduce the setup for a few minutes with one thing changed, such as a fresh battery, a different cable or a larger buffer, and see whether the dropouts disappear.

Repair options and what each can do

Recovering a dropout depends on how much audio is gone. Information that was never recorded cannot be restored; it can only be replaced or disguised.

• Patch from a backup. If a second recording exists, align it with the main one using a sharp sound, then replace each damaged section with a short crossfade. Camera audio is often poorer but complete, which is better than a gap in a key sentence.
• Remove clicks. Declicking tools in most audio editors and repair suites detect sudden discontinuities and smooth them. They work well on isolated clicks and less well on dense crackle.
• Interpolate tiny gaps. Some repair tools can fill gaps of a few milliseconds convincingly from the surrounding sound. Longer gaps cannot be filled without inventing material.
• Cut and crossfade. If the gap falls in a pause, removing a few milliseconds and crossfading may be inaudible. In a word, it will not be.
• Re-record. For a narration or a scripted piece, recording the affected sentences again is often quicker than repairing them.

Resist heavy noise reduction on call recordings with packet loss: it tends to exaggerate the warbling artefacts rather than hide them.

A worked example: an interview on a wireless lavalier

The fix for next time is just as clear: receiver closer to the guest, transmitter on the side of the body facing the receiver, and a quick walk-around test before recording.

Prevention for each kind of setup

• Wireless. Fresh or fully charged batteries, the receiver as close as practical with a clear line of sight, a frequency scan if the system has one, and the transmitter's own recording turned on if it offers internal recording as a backup.
• Computer recording. Close unnecessary apps, raise the buffer size while recording (latency matters less than when monitoring live), disable USB power saving, and plug interfaces directly into the computer rather than through a hub.
• Storage. Use cards rated for the recorder's format, format them in the device before a session, and keep plenty of free space.
• Cables. Replace any cable that crackles when moved and secure connectors so they cannot be pulled.
• Remote calls. Use a wired connection rather than Wi-Fi where possible and ask each participant to record locally, as described in setting up a remote interview recording.
• Every setup. Match sample rates across devices and software, and do a short test recording before the real one. The routine in checking audio quality takes five minutes and catches most of these problems.

How dropouts affect a transcript

Speech recognition can only transcribe what was recorded. A gap in a word usually produces a missing or wrong word; a gap across a phrase produces a sentence that jumps; robotic packet-loss patches often come out as plausible but incorrect words. Clicks are generally ignored, though dense crackle lowers accuracy around it. A stutter can produce a repeated word.

Longer silent stretches can be a risk for any Whisper-style model, which may occasionally produce text that was never said in silence or noise; Whisper hallucinations explains why and what to look for. The broader problem of skipped passages has its own guide to missing words in transcriptions. Mark known dropouts as unclear in the transcript rather than guessing what was said, especially in interviews and research material.

Limits of dropout repair

• Lost audio is gone. Repair can replace it from another source or disguise a very short gap, not recover the original words.
• Patches change the sound. A section from a camera microphone will sound different from a lavalier, which is acceptable in an interview and distracting in a polished production.
• Restoration tools that fill gaps generate plausible sound, not the actual speech, so never use filled material as evidence of what someone said.
• Fixing one dropout does not fix the cause; if the pattern is regular, the next recording will have it too.

What mydubly does with recordings that have dropouts

mydubly accepts MP3, WAV, M4A, AAC, OGG and FLAC audio and MP4, MOV, WebM, MKV and M4V video up to 2 hours per file. The browser decodes the audio, downmixes it to mono at 16 kHz and uploads it in compressed chunks of roughly 30 seconds. It does not repair dropouts, so patch the recording first if the missing words matter.

In the default deployment, recognition uses faster-whisper's voice activity filter, which skips silence while keeping timestamps tied to the original audio. A long gap therefore shows up as a jump in the timestamped transcript rather than shifting everything after it, which makes gaps easy to locate. Only the default audio track of a multi-track file is used, so if your backup recording sits on a second track, mix the two into one before uploading.

Next step: patch, then transcribe

Repair or patch the sections that matter, then run the file through audio to text and use the timestamps to check each patched passage. For interviews recorded regularly, the interview transcription page covers the rest of the workflow. A 45-minute interview costs 45 credits (4.5¢) to transcribe.