Louder is easy. Louder without distortion is the hard part
Any volume booster can multiply the sound by a number. The trouble starts at the top: a digital file has a hard ceiling, and every peak pushed past it is cut flat. That flattening is the crackle and harshness people hear when a voice memo is “boosted 300%”. Most online boosters give you a slider and leave you to listen for it.
This page does three things instead. It measures how loud the file really is, in LUFS, the loudness unit that Spotify, YouTube, Apple Podcasts and broadcasters use. It raises the whole file to a standard level. And a limiter looks 5 milliseconds ahead and turns down only the few peaks that would hit the ceiling, so they are held just under it instead of being cut. Then the new file is decoded and measured again, and the page tells you the loudness, the loudest peak and how many samples reach full scale.
Which setting to choose
- Voice & podcasts (−16 LUFS): the level Apple and most podcast hosts recommend for spoken audio. Right for voice memos, lectures, interviews, audiobook chapters and WhatsApp voice notes.
- Music (−14 LUFS): about where Spotify and YouTube play music. Louder than that, and the services turn it down again.
- By a set amount: +1 to +30 dB, with the same limiter. For when you know you want, say, 6 dB more.
- Lossless (MP3 only): no re-encoding at all; see below.
Mono files, such as most phone voice memos, are set 3 dB lower (−19 LUFS). A mono file plays on both speakers, so it sounds as loud as a stereo file 3 dB higher. Podcast loudness guides use the same pair of numbers: −16 LUFS for stereo, −19 for mono.
Why “normalize” often does almost nothing
Normalizing in most editors means peak normalization: the file is raised until its single loudest moment touches the ceiling. One cough, clap or bump of the microphone decides how far everything else can go. We made a test recording of quiet speech with one hand clap in the middle and measured both approaches:
| Quiet speech + one clap | Loudness | Loudest peak | Speech level |
|---|---|---|---|
| Original | −28.8 LUFS | −0.4 dBFS (the clap) | −34.8 dBFS |
| Peak normalization | −28.4 LUFS | 0.0 dBFS | +0.4 dB, barely changed |
| This page, −16 LUFS | −16.0 LUFS | −1.7 dBFS | +16.0 dB louder |
The limiter turned the clap down by up to 17 dB for a fraction of a second and left the speech alone. No sample of the result reaches full scale, checked with a separate loudness meter (FFmpeg’s EBU R128 filter), which agreed with this page’s own meter to within 0.2 LU on every file we tried. Because a short loud event also counts in the measurement, the page checks the result and, if limiting left it under the target, raises it once more by the difference.
Lossless MP3 boost: how it works
An MP3 is made of frames of about 26 milliseconds, and each frame stores a global gain number that scales everything in it. One step of that number is exactly 1.5 dB (a factor of 21/4). The lossless mode adds the same number of steps to every frame and changes nothing else, the method of the classic MP3Gain program. Nothing is decoded or encoded again, so there is no generation loss and the file stays the same size, byte for byte.
We tested it on a stereo 128 kbps MP3 and on a mono 24 kHz voice-memo style MP3 (which stores the number in a different place). Decoded with an independent decoder, the results were the original multiplied by exactly 213/4 and 214/4 (+19.6 and +21.1 dB), with a difference of about one part in a million, and every one of the 335 and 252 frames had moved by the same step.
The catch: the boost applies to the peaks too, and there is no limiter. The page offers the most steps the file can take before its loudest peak would clip, and if the MP3 already touches the ceiling anywhere, as in the clap recording above, it says a lossless boost is not possible.
What we measured
| File | Setting | Result |
|---|---|---|
| Quiet voice, MP3 128 kbps, −28.9 LUFS | Voice | −16.3 LUFS, peak −7.9 dBFS, same length to the sample, no time shift |
| Same, WAV 16-bit | Voice | −15.9 LUFS; every sample exactly the original × 4.22 (within half a step of 16-bit) |
| 5-minute talk, MP3 | Voice | 9 s to boost, encode and measure the result; −16.4 LUFS, exactly 5:00 long |
| Music already at −7 LUFS | +12 dB | peaks held at −1.4 dBFS, nothing clipped |
| MP3, stereo and 24 kHz mono | Lossless | +19.6 / +21.1 dB, file size unchanged, 0.03 to 0.07 s |
Formats and quality
MP3, M4A (AAC), WAV, FLAC, OGG and Opus are accepted. One exception for now: iPhone voice memos recorded with the Lossless setting are Apple Lossless (ALAC), which the browser library this page uses cannot decode yet; the default Compressed memos work. MP3 is saved as MP3 at its own bit rate and M4A as M4A; WAV and FLAC become WAV, so nothing lossless is compressed by surprise. Re-encoding an MP3 costs a little quality, as every MP3 encode does; at the original bit rate it is hard to hear, and the lossless mode avoids it entirely. The length stays exact: we removed the encoder delay that a browser’s MP3 decoder leaves at the start (1,105 samples in our test file), so the louder file lines up with the original to the sample.
Very quiet recordings have their background hiss raised along with the voice; boosting cannot tell them apart. To add the louder recording to a video, use Add Music to Video; to make an MP3 smaller afterwards, Compress MP3; to cut silence at the start or end, Trim Audio; for a WhatsApp voice note, OGG to MP3 first if an app needs MP3. Nothing on this page is uploaded: the file is measured and processed in your browser.