Dynamic Range Test — Crest Factor

A dynamic range test measures how far a track’s sustained level sits below its loudest peaks, reported as a DR score. This page decodes local audio files with the Web Audio API and runs the DR meter algorithm on 3-second blocks entirely in your browser, so nothing is uploaded.

How to use it

1. Load a track. Drag an MP3, WAV, FLAC, AAC or OGG file onto the drop area, or click to open the file picker; decoding runs locally through decodeAudioData.
2. Use a built-in preset if you have no file. Acoustic (DR14+), Brickwall (DR5) and Sine Tone (DR3) are included, and the acoustic preset loads automatically once the page is idle.
3. Read the Official DR badge. Green means DR12 and above, amber covers DR8 to DR11, and red marks DR7 and below.
4. Check the per-channel table. Left and Right rows list peak level, top-20% RMS and channel DR in dBFS. A 1 dB channel imbalance splits the two DR values.
5. Audition the file. Play Audio Preview routes decoded audio through the browser engine, and Stop ends it immediately.
6. Read the 3-second timeline. The canvas plots per-block peak and RMS against a 0 to -40 dBFS grid.

What 96.33 dB actually means

Sixteen-bit PCM is quoted as 96.33 dB of dynamic range, which is 20 x log10(65536), or 6.02 dB per bit.

That 6.02 dB is 20 x log10(2) = 6.0206 dB, so 24 bits gives 144.49 dB and 8 bits gives 48.16 dB. It is a level-counting convention: the largest code against the smallest step. A second convention measures a full-scale sine against the RMS error of an ideal rounding quantiser and gives 6.02N + 1.76 dB, or 98.09 dB at 16 bits. The 1.76 dB is the sine’s 3.01 dB crest factor plus the rounding error’s 4.77 dB advantage over half an LSB.

Real hardware falls short of both figures: analogue stages add thermal noise, clock jitter smears sampling instants, and converter linearity is imperfect. A well-built 16-bit chain delivers roughly 90 to 96 dB of A-weighted SNR, while 24-bit interfaces measure about 105 to 120 dB, around 18 to 20 effective bits. Dither and noise shaping shift the perceptual floor again: flat TPDF dither costs about 4.8 dB of broadband noise, and shaping moves that noise above 15 kHz, as the dithering demo shows.

DR conventions: peak, RMS, loudness and true peak

A DR score is a crest-factor-style statistic: it compares a track’s peak sample with the RMS energy of its loudest 3-second blocks.

This page implements the official DR meter method: non-overlapping 3.0-second blocks per channel, a peak and an RMS value per block, blocks sorted by RMS, the loudest 20% averaged into one root-mean-square figure, and DR = 20 x log10(overall peak / top-20% RMS), rounded to the nearest integer and averaged across channels. A four-minute track holds 80 blocks, so the top 20% is the loudest 16.

Other meters answer different questions:

Metric	What is compared	Unit	Question it answers
Sample peak	Loudest sample vs full scale	dBFS	Does the file hit the digital ceiling?
Crest factor	Peak vs RMS of the same window	dB	How peaked is the waveform?
DR score	Peak vs loudest top-20% block RMS	DR units (dB)	How squashed is the master?
Integrated loudness	K-weighted, gated programme loudness	LUFS	How loud after normalisation?
Loudness range	Spread of short-term loudness	LU	How much contrast survives?
True peak	Oversampled peak level	dBTP	Can a converter clip between samples?

EBU R 128 sets a broadcast target of -23 LUFS with a -1 dBTP ceiling, and streaming services normalise playback near -14 LUFS, so a hotter master no longer plays louder there. The DR score still describes a recording’s internal contrast rather than its delivery level.

What a brickwalled master looks like

A heavily limited master shows block RMS values clustered a few dB under full scale, which collapses the DR score into the DR4 to DR7 region.

On the timeline the peak and RMS traces run close together and nearly flat: transients are shaved and sustained elements are lifted by makeup gain. CD masters peak at -0.1 to -0.3 dBFS, or at the -1 dBTP ceiling used for lossy delivery. Compare two editions of one album: many 1980s CD transfers score DR12 to DR16, while hot 2000s remasters score DR5 to DR8.

Accuracy, limits and troubleshooting

• DR uses 3-second blocks, so faster events are invisible: a 200 ms cymbal contributes to its block’s peak but never gets its own reading. Short tracks give coarse results, because under about 15 seconds there are fewer than five blocks.
• The final partial block counts only when it is close to a full 3 seconds, so the tail of a long fade-out can be dropped and the DR value nudged upward.
• Mono files show one row labelled Channel Left, and lossy codecs shift peaks by fractions of a dB, so MP3, AAC and Opus versions of one master can differ from the WAV by a DR unit.
• Values are sample peaks in dBFS, not oversampled true peaks in dBTP: a file reading -0.1 dBFS can still clip inside a consumer DAC.
• OS audio processing colours the preview only; the DR number comes from decoded samples, not the system mixer.
• DR is a descriptive index, not a verdict on quality. A DR5 master can be exactly right for the music.

Related tools

• Blind Dynamic Range Test — check by ear whether you can hear compression once playback levels are matched.
• Loudness Meter — read integrated loudness and true-peak values alongside the DR score.
• Spectrum Analyzer — see the distortion products and noise floor that limiting leaves behind.