Audio repair samples
Hear what each repair changes before you trust it with your own recording. These 16 before/after pairs cover eight fixes, including cases that stay damaged or are left alone. Download either side to compare in your editor. No account needed.
Most clips are matched to the same loudness so louder does not win the comparison. The quiet-audio clips preserve the volume change. Start at a comfortable listening level; nothing plays automatically.
Small room
A software room that rings for 0.55 s. The room tail falls. Echo repair is beta; it cannot recreate a recording made in a dry room.
12 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Measured room decay:
0.52 s → 0.28 s
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Big room, noisy
The same 1.4 s room with hiss 12 dB under the voice. The noise masks the room, so the echo check is unsure and this repair leaves the file alone. The full evidence page also tries noise removal first.
12 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Measured room decay:
not measured → not measured
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
A fan
Pink noise 10 dB under the voice. The steady background falls. Compare the consonants as well as the pauses.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Speech above the stillest background:
17 dB → 65 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Someone else talking
A second reader 12 dB under the voice. The second speaker remains audible. Noise reduction is designed to preserve speech, including the voice you wanted to remove.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Speech above the stillest background:
53 dB → 62 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Dull, 3 kHz
Low-passed at 3 kHz. More of the voice's top end comes through. Listen for clarity rather than a volume increase.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3High-frequency energy relative to the voice body:
-54.7 dB → -20.9 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Dark, 1.5 kHz
Low-passed at 1.5 kHz. The top end improves but still measures severely muffled. EQ cannot replace detail that the filter removed.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3High-frequency energy relative to the voice body:
-78.7 dB → -62.7 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Distorted
Measurements and more examplesClipped in the recorder
Pushed 9 dB over full scale and cut off flat. Flattened peaks are rebuilt and the waveform gets closer to the original. The damaged take is still an estimate, not the lost original.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Samples detected as clipped:
1.2 % → 0 %
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Overdriven before the recorder
Pushed 9 dB into a soft saturator, so the tops are bent round instead of cut off. Soft saturation is not flagged as hard clipping. The repair leaves these bent peaks alone.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Samples detected as clipped:
0 % → 0 %
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Too quiet
Measurements and more examplesQuiet take
The reading turned down to -36 LUFS. The voice gets louder, with peak control. Start at a comfortable listening volume before switching to the result.
10 SEC · MP3 · VOLUME CHANGE PRESERVED
before
Download before MP3after
Download after MP3Integrated loudness:
-36 LUFS → -16.5 LUFS
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Quiet take with hiss
The same take with hiss 30 dB under the voice. Turning up the voice turns up the hiss too. Noise removal is a separate step; this comparison uses only the level fix.
10 SEC · MP3 · VOLUME CHANGE PRESERVED
before
Download before MP3after
Download after MP3Integrated loudness:
-36 LUFS → -16.5 LUFS
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
One side dead
The voice on the left, nothing at all on the right. The live left channel is copied to both ears. Listen in stereo to hear the change.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Gap between the two channels during speech:
96.4 dB → 0 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Two people, one each side
One voice on the left, a second voice 10 dB down on the right. Both voices are mixed into both ears. Their relative levels stay the same; balancing channels does not make the quiet guest equally loud.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Gap between the two channels during speech:
8.4 dB → 0 dB
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Mains hum
60 Hz and its first 4 harmonics, 22 dB under the voice. The low pitched tones fall. Compare the pauses and the low body of the voice.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Detected mains tones relative to speech:
-23.2 dB → none detected
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
A buzz
60 Hz with 24 harmonics, 22 dB under the voice. The first five tones fall but the bright buzz above them remains. A fine hum check does not mean every harmonic is gone.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Detected mains tones relative to speech:
-24 dB → none detected
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Gusty wind
Low rumble that swells and drops, 22 dB under the voice. The gusting low rumble falls. Listen for the low body of the voice that the filter should keep.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Time detected as windy:
19.7 % → 0 %
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Plosives
8 low thumps of 80 ms at word starts, as loud as the voice. Short plosive thumps are not detected as sustained gusts. This wind repair leaves them alone.
10 SEC · MP3 · MATCHED TO -23 LUFS
before
Download before MP3after
Download after MP3Time detected as windy:
0 % → 0 %
One repair at 100% amount. No other repair in the chain. Readings are before playback level matching and MP3 encoding.
Where these come from
Two public-domain LibriVox readings from Short Story Collection 107 are our clean sources. Voice A starts 40 seconds into track 3; voice B starts 120 seconds into track 4. We add the damage in software, so these are controlled examples, not recordings from real fans, microphones or rooms. They show behavior on these voices, not a promise for every recording.
Each result uses one repair at 100% amount. The working tool starts at 80%, so its first result may differ. The MP3 listening files are encoded at 192 kbps. The measurements come from decoded PCM before playback level matching and encoding; MP3 can change peaks and other readings. A missing reading means the check could not judge it, not a perfect score.
How we test explains the checks and fixture tests that verify these files and numbers. The sample manifest lists sources, recipes, settings, measurements and both download URLs. For more comparisons, follow each repair's evidence link above.