MP3 Volume Booster

It starts the same way every time. Track one is a rip from a 1990s CD, mixed quiet and proud. Track two is a voice memo you recorded from the back of a lecture hall, so faint that you crank the volume to hear the professor breathe. Track three is a modern master, slammed to the ceiling, and it detonates. Your thumb does the equalization by hand — up, up, down, down — for the length of the commute.

The problem isn't your headphones or your player's weak amplifier. It's that "volume" in digital audio is a property of the file itself, decided at recording or mastering time, and the files in your library were decided by different people with different philosophies. A lecture captured at a polite distance and a war-era pop master don't disagree about anything except how loud they should be — and neither one will ever meet the other halfway on its own.

This tool settles it in one pass. Load a quiet file and boost it by decibels with clipping protection; load a folder's worth of files and normalize them all to one measured loudness target so shuffle stops being a lottery. The whole job runs inside your browser — decode, measure, gain, limit, re-encode — with an A/B preview so you can hear exactly what you're committing to before you commit to it.

1. 1

   Load the MP3 (or the batch)

   Drop one file for a boost, or several for a normalization run. Each is decoded to raw samples locally — MP3, M4A, WAV, whatever your browser can read, all welcome.

2. 2

   Read the loudness report

   The tool measures integrated loudness in LUFS, true-peak level in dBTP, and dynamic range before touching anything, so you know why the file sounds the way it does.

3. 3

   Choose boost or normalize

   Boost applies your dB value to one file. Normalize computes per-file gains toward a target — -16 LUFS for speech, -14 for streaming-parity music, or any number you specify.

4. 4

   Set the safety ceiling

   The default true-peak limit of -1.0 dBTP keeps boosts out of clipping territory, including the inter-sample peaks that simple peak meters miss.

5. 5

   A/B before you commit

   Toggle between original and processed. If the limiter is working hard you'll hear it in the preview — punch turning to flatness — and you can back the boost off a decibel or two.

6. 6

   Export

   MP3 at 256-320 kbps for everyday use, or WAV when this is the archival copy. Batch runs write files with matched loudness and a one-click download-all.

Digital audio has a mathematically hard ceiling: 0 dBFS, full scale, the value of the largest sample the format can express. Anything past it doesn't get louder — it gets flattened, with the tops of waveforms sheared into rectangles that read as harsh, buzzy distortion. Everyone who records audio therefore leaves headroom, a safety margin below the ceiling, because a singer who laughs louder than the soundcheck will clip a track recorded at the limit.

Recording engineers aim their margins conservatively. A podcast or lecture captured with 8 to 16 dB of headroom is a recording that can never be surprised; it's also, played back at a normal volume setting, a whisper. Field recordings made on phones lean quiet for the same reason, plus distance: every meter between mouth and microphone costs real level. None of this is a defect. It's caution, stored as dB.

Mastering engineers made the opposite trade for two decades. The loudness war — the arms race of the CD era — pushed commercial masters toward the ceiling so songs would sound bigger on radio and shuffle. Those files sit at -7 to -9 LUFS, which is why a 2005 pop master plays roughly twice as loud as your lecture. Two philosophies, one library, and a volume knob caught in the crossfire.

A note on perception while we're here: loudness grows on a logarithmic scale, so it takes about 10 dB of gain for something to sound "twice as loud." A boost of +3 dB is barely a nudge; +6 dB is clearly louder; +10 dB is a different experience. When you're deciding how much to ask for, that's the ruler to think in.

The old approach — peak normalization — scales a file so its single loudest sample hits a chosen level, say -1 dBFS. It's cheap and it's honest, but it measures the wrong thing. A jazz recording with one cymbal crash and a wall-of-sound master can both normalize to the same peak while sounding nothing alike, because the crash occupies a hundredth of a second and the wall occupies all of it.

Loudness normalization measures what a human actually perceives. The standard method, defined in ITU-R BS.1770 and adopted by the EBU's R128 recommendation, runs the audio through a filter shaped like human hearing, averages it over time with gating that ignores silence, and produces a single number: integrated loudness in LUFS. Two files at the same LUFS sound equally loud, regardless of where their peaks live. This is the same measurement every streaming service uses to decide playback level.

The consequences are practical. Peak normalization of a lecture with one cough in it barely changes anything — the cough is the peak, and everything quieter stays quiet. Loudness normalization brings the speech itself to target, tucking the cough under with the limiter if it crosses the ceiling. For anything with a voice in it, LUFS is not the fancy option; it's the correct one.

Say your lecture measures -24 LUFS and you boost it 8 dB to reach -16. The peaks, which sat at -8 dBFS, now sit at 0 — the exact edge of the cliff, where true-peak overshoots will clip even though the sample meter reads clean. This is where the booster stops being a multiply-and-hope tool. A true-peak limiter watches the boosted signal and pulls down only the moments that would cross the ceiling, a few milliseconds at a time, with enough release smoothing that the correction is inaudible.

True peak matters because of a subtlety in digital audio: the reconstructed signal between samples can exceed every sample value. A file that meters at -0.3 dBFS can reconstruct above 0 dBFS after the DAC does its work, which is why mastering engineers set ceilings at -1.0 dBTP and why the default here does the same. Simple peak checking misses inter-sample overshoot; true-peak estimation, standardized in BS.1770's later revisions, catches it.

There's an honest limit to what a limiter can do for you. It's a tool for the exceptional moment — the laugh, the dropped microphone, the cough — not a tool for squeezing 10 dB of extra loudness out of an already-loud master. Push a limiter that hard and it stops being insurance and becomes compression with no musical judgment: drums lose their snap, speech acquires a strained edge, and the file sounds louder only in the way a shouted sentence is louder than a spoken one. The A/B preview exists so you can hear that trade and decline it.

The batch line is the quiet hero. Gains ran in both directions — three war-era episodes came down 5 to 7 dB, eleven modest recordings went up 2 to 10 dB — and afterward, shuffle across all fourteen files holds one comfortable volume setting. That's what -16 LUFS means in practice: an agreed meeting point, not a maximum.

Notice File A's limiter figure: 0.7% of samples touched the ceiling. That's the limiter doing precisely its job — catching the lecture's two coughs and one dropped binder without leaving a fingerprint on the other 99.3% of the recording. If that number had read 15%, the honest response would be a smaller boost, and the tool would say so in the report.

Veterans of the MP3 era know a third road: ReplayGain, the classic open standard that measures a file's loudness and stores the result as a tag, letting capable players apply the gain at playback time without rewriting audio. It's elegant — the file is never re-encoded — and it's why foobar2000 and VLC libraries could sound level-matched twenty years ago.

Its weakness is the ecosystem. Tag-based leveling works only in players that volunteer to read the tag, which rules out every car stereo, most phone apps, half the smart speakers on the market, and anything that plays files through a platform's own normalization. A gain baked into the samples travels with the file into every context there is.

This tool bakes. The measurement method is the same lineage — BS.1770-style filtering, integrated LUFS — but the correction is written into the audio at export, with the limiter as safety. If your entire listening life happens inside one ReplayGain-aware player, the tag route remains respectable. If your files ever leave that player, baked gain is the one that survives the trip.

# Single file to a spoken-content target
ffmpeg -i in.mp3 -af loudnorm=I=-16:TP=-1.5:LRA=11 -b:a 256k out.mp3

# Plain +6 dB boost, no measurement
ffmpeg -i in.mp3 -af volume=6dB -b:a 256k out.mp3

• Already-loud masters. A -8 LUFS pop file can absolutely be brought toward -14 — normalization lowers it, which is fine — but boosting it further engages the limiter continuously and hollows out the sound. If a file measures louder than your target, the correct gain is negative, and the tool will simply apply it.
• Wide-dynamics classical and jazz. A recording with 30 dB between pianissimo and fortissimo normalized to -14 LUFS spends much of its life below the level you'll want, and the quiet passages were quiet on purpose. Consider a smaller target shift, or boost by ear instead of by number.
• Noise floors come along. Gain raises the hiss, the HVAC, and the café exactly as much as it raises the voice. Recordings made in bad rooms often need cleanup first — trimming dead air and gaps with the Audio Silence Remover is the cheap pass that makes any boost sound better.
• Clipping baked into the source. If the recording was clipped at capture, the flat-topped samples are already deformed; no boost or limiter repairs them, and boosting makes the artifact more audible. The report flags suspicious flat-top density so you know the damage predates this page.
• Very long files. An MP3 decodes to memory for measurement and gain; a three-hour file is fine on a laptop and miserable on an old phone. Splitting first, or processing in parts, keeps things comfortable.
• Meta-loudness on streaming platforms. If you target -14 LUFS and upload to a service that normalizes to -14, the platform may still turn your file down when its measurement disagrees with yours by a fraction. Aim for the target, accept the last half-decibel as the platform's opinion.

• Podcasters delivering episodes to a -16 LUFS spec, or inheriting guest recordings that arrive at -26 and need to join the rest of the catalog.
• Musicians checking demo mixes against streaming loudness — hearing your master at -14 LUFS is the fastest reality check on how it will sit next to the catalog it joins.
• Students and commuters amplifying lecture captures and conference recordings whose quietness was baked in at the back of the hall.
• Audiobook and sermon archivists leveling decades of irregular recordings into one listenable collection.
• Aerobics instructors and event DJs matching mixed-era playlists so one fader position survives the whole session.
• Field researchers and journalists whose interview levels vary with every room, mic placement, and passing truck.
• Anyone who recorded into a browser and got a clean but timid file — record with the Online Voice Recorder, then bring the result here to make it sit properly.

It's also the second step of a standard cleanup chain. Convert iPhone voice memos to MP3 first with the M4A to MP3 Converter, cut the silence, then level what remains. And when the question is 'what frequency is that hum?' rather than 'why is this quiet,' the Tone Generator answers it in a different but neighboring way.

If there's a philosophy here, it's that loudness decisions are content decisions. A -16 LUFS spoken file and a -9 LUFS dance track are both legitimate; the failure mode is when levels are accidents — inherited, unmeasured, and unlistened-to. Measuring first, then deciding, turns an accident into a choice, and choices are what editors are for.

The files people want to make louder are rarely public: voice memos, lectures, interviews, rough mixes, family recordings. The classic online fixer asks you to ship exactly those files to a queue, wait, and download the result. This page runs the entire pipeline — decode, BS.1770-style measurement, gain, limiting, re-encode — in your browser, and the audio never crosses the network at any point in the process.

Batch work benefits twice. Nothing uploads, so a folder of 40 episodes processes as fast as your machine decodes them rather than as fast as your connection permits; and nothing persists, so there's no library of your voice sitting in someone else's storage tier after the tab closes. Load the page, work offline, close the tab, and the only trace of the session is the louder files you chose to keep.

Volume is one link in a chain, and the rest of the chain lives nearby: cut dead air before leveling with the Audio Silence Remover, convert iPhone voice memos with the M4A to MP3 Converter, diagnose mystery hums with the Tone Generator, and capture new audio straight into the browser with the Online Voice Recorder. Loud files start as good recordings — these are the tools that make them both.