Fix uneven audio levels in a recording

Save an untouched original, then open a working copy. If the recording may be used for a case or another consequential purpose, keep editing attempts separate from the source. A listening improvement does not establish what a disputed word says.

Choose a quiet sentence and a loud sentence from the same recording. Listen to both at a comfortable playback volume. Is one entire speaker quieter, or does the same speaker change level throughout an answer? Do the loud passages already sound harsh or crackly? Note those differences before choosing an effect.

If you use an online tool, check where it processes and stores files before sending sensitive material. A convenient upload button does not tell you how the recording will be handled.

For a small number of changes, adjust only the affected passage. Clip gain changes the level of a selected clip or region. A volume envelope, sometimes called volume automation, lets you draw changes over time. Audacity's Envelope Tool documentation describes this approach.

Lower an unusually loud answer or gently raise a quiet one, then listen across the boundaries. Make the transition gradual enough to avoid an obvious jump. Keep surrounding words and pauses in the comparison so the change fits the conversation.

Adding gain raises everything in that selection, including hiss, room noise and other voices. If the wanted speaker remains buried, more gain may make the whole passage louder without making it easier to understand. Watch the peak meter as well as listening; a quiet passage can still contain a brief loud sound.

• Local gain or an envelope can target a passage or change over time.
• Peak normalization applies one gain change across the selection.
• Downward compression reduces sound above its threshold.
• Limiting controls peaks above its ceiling.

Source: Audacity, Normalize; Envelope Tool; compression and limiting guide. Shapes show scope or conditions, not improvement or a processing preset.

A compressor can narrow the dynamic range: the difference between louder and quieter parts. A typical downward compressor turns down sound above a threshold. Makeup gain can then raise the processed signal. The quiet passages become closer in level to the controlled loud passages; they are not magically separated from the background.

Start with a short section that includes both. The main controls answer different questions:

• Threshold: Which louder sounds should trigger reduction?
• Ratio: How much reduction happens above that threshold?
• Attack and release: How quickly does reduction engage and let go?
• Makeup or output gain: How loud is the result after reduction?

Use the smallest amount that makes the conversation easier to follow. Listen for flattened emphasis, dull consonants, or noise that seems to swell between words. Ease the processing if those changes become distracting. If the compressor never engages, its threshold may be above the passages you are trying to control.

No threshold, ratio or timing preset fits every recording. A setting demonstrated on a clean voiceover may work poorly on a distant interview with background noise. Audacity's compression and limiting guide explains the controls and the role of makeup gain.

• Input enters the compressor reduction stage.
• Threshold determines where reduction applies; ratio determines reduction above threshold.
• Optional makeup gain raises the processed output level after reduction.

Source: Audacity, compression and limiting guide. Makeup gain is one option; normalization can be a separate later adjustment. This is not a universal processing chain.

Once the passages sit together more comfortably, decide whether the overall level needs adjustment.

Peak normalization applies one gain change so the highest peak reaches a chosen level. The quiet and loud passages keep the same relative difference. If one sudden sound already approaches the peak limit, normalization may barely raise the rest. It is not a substitute for leveling those passages first. The Audacity Normalize manual describes that single gain change.

Loudness normalization uses an average or perceived loudness measurement rather than just the highest peak. It can help match the overall listening level of separate clips. Applying a single loudness adjustment to a whole recording still does not necessarily resolve variation between its sentences.

If you must deliver to a particular platform, broadcaster or recipient, use its actual specification and check both loudness and peaks. There is no universal LUFS target for every recording. Leave headroom rather than trying to fill every available bit of the meter.

A limiter controls peaks above a ceiling. It can be useful near the end of processing, but it should not have to fight the entire conversation. If it is working heavily or the voice sounds squeezed or distorted, lower the level or revisit the earlier gain and compression changes.

Limiting new peaks does not undo clipping already present in the source. If the original is harsh on loud syllables, treat that as a separate repair question rather than assuming a lower ceiling will recreate the missing detail.

Before keeping the result:

1. Return to the same passage. Compare the untouched source and the processed version, including a quiet answer, a loud answer and a pause.
2. Use similar comfortable listening levels. A louder version can seem more useful simply because it is louder. Listen for words and natural speech detail as well as volume.
3. Check what became distracting. Listen for raised background noise, breathing, abrupt level changes, pumping or harsh consonants. Reduce or undo changes that make the passage less useful.
4. Open the exported file separately. Check it outside the editor and inspect its peaks if a meter is available. Confirm that the export contains the intended version and that video remains in sync when applicable.

Keep the project, the original and the chosen export clearly labeled. Avoid repeatedly making lossy exports into the input for the next attempt when a higher-quality working copy is available.

• Keep the original unchanged as the reference.
• Try a change on a working copy.
• Compare the same passage at similar listening levels.
• If less useful, reduce or undo and compare again.
• Export the chosen version, then open that export separately.

Source: the guide’s Keep the original and Compare the passage sections. Solid arrows show the defined workflow; the dashed return shows a reversible revision. No completion state is implied.

Level adjustment helps when useful speech was captured at an awkward volume. It cannot guarantee words that remain masked by noise, overlapping voices or distortion. Stop increasing the effect if the background becomes dominant or the conversation sounds less natural.

If the passage still matters and you cannot make it easier to follow, an engineer can assess the recording itself. Media Medic's audio forensic service includes restoration work, and an audio engineer listens before an invoice is sent. That assessment is about what may improve in your file; it is not a promise that every word can be recovered.

Request a free file review. Tell us where the conversation becomes hard to hear. You can explain the problem before sending sensitive material.