Audio Quality Check

About Audio Quality Check

Audio Quality Check decodes a file in your browser and gives it a technical health check: sample peak, true peak, RMS, integrated loudness in LUFS, loudness range, clipping, DC offset, stereo phase, silence at the edges, and a level graph you can hover over. It then turns those numbers into plain recommendations and shows how many dB you would have to add or remove to hit common streaming targets.

It is built for the moment before you publish or process a file: a podcast episode that sounds quiet next to other shows, a mix that distorts after MP3 export, a voice memo with a long dead intro, a song that feels different on the phone speaker. Instead of guessing, you see which number is off and which tool fixes it. Nothing is uploaded — analysis runs on your device, so unreleased tracks and private recordings stay private.

One honest limit: the tool measures levels and signal health, not codec quality. It cannot tell you the bitrate of an MP3 or whether a file was upscaled from a lower-quality source.

How to check a file

What the numbers mean

• Sample peak (dBFS) — the highest stored sample, where 0 dBFS is digital full scale. Staying below about −1 dBFS leaves room for encoding.
• True peak (dBTP) — the highest level of the reconstructed waveform between samples, estimated with 4× oversampling. It can be higher than the sample peak, and it is the number lossy encoders and DACs actually react to. Keep it at or below −1 dBTP.
• RMS (dBFS) — average signal energy with no weighting. Handy for comparing the body of two files, but bass-heavy material reads higher than it sounds.
• Loudness (LUFS) — integrated loudness using K-weighting (which mimics how the ear hears frequencies) and gating (which ignores near-silent passages). This is the figure streaming platforms normalize to.
• Loudness range (LU) — the spread between the quiet and loud parts, measured over 3-second windows. Speech is usually low, orchestral music high; very large values in a podcast mean listeners will keep reaching for the volume knob.
• Crest factor (dB) — peak minus RMS. A small value means a dense, compressed signal; a large one means sharp transients over a soft body.
• Stereo correlation — +1 means both channels are identical, 0 is wide and uncorrelated, and negative values mean phase cancellation when the file is played in mono.
• DC offset — a constant shift of the waveform away from zero. It wastes headroom and can cause clicks where audio is cut.

Loudness targets cheat sheet

Platforms turn loud files down to a reference level so every track plays at a similar volume. These are the commonly cited values — services change their policies, so treat them as a guide and check the current specification when it matters.

The report's “Gain to reach platform targets” card does this arithmetic for your file: it subtracts your measured loudness from each target and warns you when the extra gain would push true peak past the ceiling.

Common problems and what to do

How the measurement works

The file is decoded at its own sample rate whenever the header can be read, so nothing is resampled before measuring. Loudness is computed per channel with the two K-weighting filters from ITU-R BS.1770-4, summed with the standard channel weights, split into 400 ms blocks that overlap by 75%, and gated: blocks below −70 LUFS are dropped, then blocks more than 10 LU under the remaining average. Loudness range uses 3-second windows with a −20 LU relative gate and compares the 10th and 95th percentiles, as in EBU Tech 3342.

Because the whole file is decoded in memory, very long recordings need a lot of RAM — a one-hour stereo file takes roughly 1.3 GB. For multi-hour files, trim a representative section first. Results describe the audio your browser decodes, so a few tenths of a dB of difference from a hardware meter, or from another decoder's handling of MP3 padding, is normal.