CD Dynamic Range Measurement Tool
  • Python 95.2%
  • Shell 4.8%
Find a file
2026-10-15 00:02:09 -04:00
tests History note: a repeat reading of one disc is not a second copy 2026-10-11 20:54:01 -04:00
.gitignore Add drcd: DR14 dynamic range meter for audio CDs and files 2026-10-11 18:09:17 -04:00
drcd Switch copyright holder and contact address to kaiser@hofburg.pub 2026-10-15 00:02:09 -04:00
install.sh Finish macOS docs, python3 prerequisite, history header 2026-10-11 19:15:06 -04:00
LICENSE Initial commit 2026-10-11 18:03:23 -04:00
README.md docs: clone URL is now git.hofburg.pub 2026-10-14 01:03:39 -04:00

drcd

Dynamic Range (DR) meter for audio CDs and audio files — measures DR on the same scale as https://dr.loudness-war.info (the DR14 / Pleasurize Music Foundation / TT Dynamic Range Meter scale), so the numbers are directly comparable with the values published on that site.

Pop a CD in the drive, run one command, read the DR number off the screen:

$ drcd
Ripping with cdparanoia … ✓ 12 tracks
  [ 1/12] 01 - Track name          DR12
  ...
  [12/12] 12 - Track name          DR9

/media/cdrom (HL-DT-ST DVDRAM GP65NB60)
   #  DR         Peak dB    RMS dB   Length  Title
  --------------------------------------------------------------------
   1  DR12         -0.20    -14.35     3:45  01 - Track name
   ...
  12  DR9          -0.09    -12.02     4:11  12 - Track name
  --------------------------------------------------------------------

  Album DR 11  (tracks 12, min 7, max 14)
  measured in 3:41
  log: /home/you/DR-logs/Artist - Album - DR.txt

The number to write down when comparing two copies is the Album DR (the same figure loudness-war.info lists in its "DR" column, with min/max as the per-track extremes).

Install

Needs python3 + numpy, ffmpeg, and a CD ripper if you want to read discs. ./install.sh installs the script into ~/.local/bin, checks the dependencies and solves the awkward one (numpy) by itself.

Linux (Arch / CachyOS / Debian / Ubuntu …)

sudo pacman -S python-numpy ffmpeg cdparanoia      # Arch / CachyOS / Manjaro
sudo apt install python3-numpy ffmpeg cdparanoia   # Debian / Ubuntu / Mint
./install.sh

Drive permissions (once): your user must be able to read /dev/sr0. On Arch add yourself to the optical group and re-login; on Debian it is cdrom:

sudo usermod -aG optical "$USER"

After pulling updates, run ./install.sh again (or invoke the copy in the repo, ./drcd): ~/.local/bin/drcd is a copy, so an updated repo does not change the installed command until you re-install it. A log's first line names the version that produced it (drcd 1.2.0 — DR14 (…)), which is the quickest way to tell whether you are running the version you think you are.

macOS

brew install ffmpeg libcdio-paranoia     # note: there is no brew cdparanoia formula
git clone ssh://git@git.hofburg.pub:2222/kaiser/drcd.git
cd drcd && ./install.sh                  # also makes a private venv holding numpy
drcd                                     # disc in the drive — no device argument

macOS specifics, all of which the tool already handles:

  • Ripping. Homebrew doesn't package cdparanoia; libcdio-paranoia provides cd-paranoia — the same Paranoia 10.2 code on top of libcdio, with the same options (-Q, -B, -w, -Z, -X) and the same printed output, so the rip and the measurement are identical. drcd looks for either name. MacPorts alternative if you prefer the Xiph build: sudo port install cdparanoia.
  • No --device needed. macOS has no /dev/sr0; drcd detects that and lets the ripper find the drive itself (cdparanoia -Q with no -d).
  • The drive. Modern Macs have none built in — use any USB CD/DVD drive. macOS shows it as an "Audio CD" and Music.app may offer to import it; that doesn't interfere, but you can stop it in Music → Settings → General → When you insert a CD.
  • python3. Apple only ships python3 once the developer tools are installed: xcode-select --install (or brew install python). Check with python3 -V before you install drcd, since the venv below is built with it.
  • numpy. macOS ships no numpy, and Homebrew's python refuses global pip installs (PEP 668, "externally managed environment"). So install.sh creates a private venv at ~/.local/share/drcd/venv with numpy in it, and drcd re-execs itself into that interpreter automatically when the python3 it started under has no numpy. Nothing to export, no --break-system-packages. (Preferred alternative: brew install python numpy, then it isn't needed.)
  • Eject. drcd --eject uses drutil tray eject on macOS, eject on Linux.
  • Apple Silicon vs Intel. Nothing to configure — whichever Homebrew prefix your PATH points at (/opt/homebrew or /usr/local) is used.

No ripper at all? You can still measure: Music.app rips bit-exactly to ALAC or AIFF with error correction enabled (Music → Settings → Files → Import Settings → "Use error correction when reading Audio CDs"), and then

drcd ~/Music/Music/Media.localized/Music/<Artist>/<Album>

measures those files — ffmpeg and numpy are the only extras it needs.

Set the import format to AIFF or Apple Lossless first: Music.app's default is AAC, which is lossy, and a DR measured from a lossy encode is not the CD's DR (it can also differ by a point from another tool's rip of the same file). Third-party rippers work just as well — XLD is the usual macOS choice and can write FLAC, which Music.app cannot.

Usage

drcd                          # rip the CD in the drive and measure it
drcd /dev/sr1                 # pick a specific drive (Linux; USB drives: check ls /dev/sr*)
drcd ~/Music/Album            # measure a folder of files (FLAC/WAV/MP3/…)
drcd track1.flac track2.flac  # measure specific files
drcd --mb                     # CD mode: MusicBrainz titles (this is the default)
drcd --no-mb                  # never contact MusicBrainz (works offline)
drcd --keep-rip ~/rips        # CD mode: also keep the ripped WAV files
drcd --history                # show every measurement so far
drcd -j                       # JSON output (for scripting)
drcd --fast                   # skip cdparanoia's error correction (quicker, riskier)
drcd --strict                 # treat a scratched/unreadable track as a failure

On macOS, omit --device entirely — the drive is autosensed.

Useful extras: --artist / --album (name the disc yourself instead of MusicBrainz), --eject, --no-log, --log FILE, --outdir DIR (default ~/DR-logs), --no-color, --resample 44100, --blocks exact.

Use --strict when you are comparing two copies and want to be sure a scratched disc can't win on a mis-read: cdparanoia normally interpolates over unreadable sectors (and the exit status warning tells you when that happened).

Log files are written in the plain-text DR log format that dr.loudness-war.info accepts for uploads ("DR/Peak/RMS/Duration/Title", Number of tracks, Official DR value). If a log with the same name already exists, the new one gets a timestamp appended instead of overwriting.

Comparing two copies of the same CD

That is what the history file is for. Run the tool on copy A, then on copy B; on the second run you get:

  NOTE: this disc was measured before:
    2026-10-11 18:22  DR9   Stevie Ray Vaughan / Texas Flood   identical
    keep the copy with the higher DR.

Copies are recognized three ways, in order of confidence:

  1. the disc's own table of contents (the MusicBrainz disc ID) — the same pressing;
  2. artist + album — another pressing of the same album, which has a different TOC and therefore a different disc ID. These names come from MusicBrainz (on by default in CD mode, see below) or from --artist / --album;
  3. the album's shape — with no metadata and no network at all: same track count (4+) and every track within 2 seconds. The note then says matched by same track layout (another pressing?) so you can see why it matched.

Real example: two copies of Texas Flood (SRV) — different pressings, so different disc IDs and, with --no-mb, no names either. They matched on layout and both measured album DR 13 (tracks 12,13,14,13,12,11,15,13,13,14), i.e. a tie: keep either.

drcd --history prints the whole log; drcd --history "pink floyd" filters it.

MusicBrainz lookup (on by default for CDs)

In CD mode drcd asks MusicBrainz for the artist, album and track titles using the disc's table of contents — so logs carry real titles and the comparison above works across pressings. Details:

  • Needs the TOC, so it can't work when the TOC itself is unreadable (it says so and carries on).
  • Rate-limited by MusicBrainz to about one request/second, so it retries the 503/429 "server is busy" responses a few times before giving up.
  • One disc ID can be attached to several releases (original pressing, reissue, SACD layer…). drcd prefers the medium that actually lists your disc ID, a plain CD over an SACD layer, a track count matching your disc, and the earliest date.
  • The outcome is printed and written to the log, so a log is self-documenting: MusicBrainz lookup: Stevie Ray Vaughan and Double Trouble — Texas Flood (1983, CD) or unavailable (HTTP 503) — no titles.
  • --no-mb disables it entirely (nothing leaves the machine); --artist / --album override or supply the names yourself.

Empty and silent files are not tracks

cdparanoia's batch mode sometimes writes a bogus track00.cdda.wav before track 1 — documented behavior, and it depends on the drive, not the disc (usually a fraction of a second of silence, sometimes a few seconds of it). Digital silence scores DR 0, so left alone it drags the album average down and makes two copies of the same album look different.

drcd therefore ignores any ripped file that is shorter than 0.5 s or that contains nothing but silence: it stays out of the table and out of the album DR. The runs that produce one say so:

  [ 9/ 9] track 00                                      ignored — empty file (0.1 s)
  ...
  Album DR 12  (tracks 8, min 11, max 14, 1 empty/silent ignored)
  ignored: track 00 — empty file (0.1 s) (not counted in the album DR)

and the log records the same under Ignored:, so nothing is silently dropped. Titles are keyed by the disc's own track number (taken from the file name), so a phantom file can't shift every title by one either.

What the numbers mean

  • DR — for each track, the level difference in dB between the track's near-highest peak and the RMS level of its loudest 20% (in 3-second blocks), averaged over the channels. Higher = more dynamics preserved.
  • Peak dB — highest sample level (0 dB = digital full scale). over means the file contains samples above full scale (clipping/intersample peaks); the master is probably clipped.
  • RMS dB — average level of the whole track.
  • Rough guide used on loudness-war.info: for rock/metal, DR 8+ is considered fine; DR 5–7 is heavily compressed; DR 3–4 is a brick; classical/jazz can reach DR 14–20.

A CD rip is bit-exact, so the DR you measure from a disc is the DR of that pressing. Two pressings of the "same" album (or a CD vs. a later remaster) often differ by several DR points — that is the whole point of this tool.

Algorithm (and why the numbers should match the site)

This is a faithful re-implementation of the published DR14 procedure used by the TT DR Offline Meter (the tool behind the DR database) and by dr14_t.meter (the reference Linux implementation):

  1. Split the track into blocks of 3 seconds (non-overlapping).
  2. Per block and channel: peak = max|x|, rms = sqrt(2*Σx²/N) — the √2 makes a full-scale sine read 0 dB on both, which is the meter's internal convention.
  3. Sort blocks by RMS; take the loudest K = max(1, floor(0.2*M)) blocks and average their power: R = sqrt(Σ rms²/K).
  4. Reference peak = the second-highest block peak (skipping the single highest makes it robust against one isolated click).
  5. DR_channel = 20*log10(reference_peak / R); silent or out-of-range channels count as 0; the track's DR is the mean over channels, rounded.
  6. Track peak/RMS are the true maxima/averages over the file; the album ("Official") DR is the rounded mean of the track DRs, exactly like the reference tools.

Deliberate quirks kept for compatibility: at 44.1 kHz the classic meter uses 132480-sample blocks (3.0041 s), i.e. 3×(44100+60) — this is what --blocks old (the default) reproduces. --blocks exact uses plain 3*fs blocks, which is what the Windows meter effectively measures after it resamples to 44.1 kHz; on real music the two usually agree to the same integer, and in our tests they never disagreed.

Validation

Measured with this script, compared against the reference tools and against values published on dr.loudness-war.info:

Material Reference Reference result drcd result
5 synthetic signals (sine, decaying bursts, limited mix, quiet/loud noise) dr14_t.meter 2.1.1 DR 0/0/2/10/14, peaks & RMS to 0.01 dB identical
Nine Inch Nails, Ghosts V tracks 01/02/05, official MP3 download published on dr.loudness-war.info (log by Dr14 T.meter 1.0.16) DR 8 / 9 / 5, peaks −0.00 / −0.21 / −0.09 dB DR 8 / 9 / 5, same peaks
20-minute 44.1 kHz file dr14_t.meter 2.1.1 DR 10, peak −4.26 dB, RMS −18.14 dB identical
6-track "CD" (44.1 kHz 16-bit) dr14_t.meter 2.1.1 DR 14/2/10 each, peaks/RMS identical identical
same material, resampled to 44.1 kHz simple-dr-meter (independent implementation, loudness-war-compliant) DR 8 / 9 / 5 DR 8 / 9 / 5

Other checks: 74-minute CD-length file measured in 24 s with 57 MB peak RSS (the file is streamed, never fully loaded); a silent track yields DR 0 instead of a crash; a 0.02 s file is handled; a MusicBrainz disc ID computed from a cdparanoia -Q table of contents matches MusicBrainz's own published test vector.

Two test files ship with it:

Test Covers
tests/test_drcd_core.py disc-ID vector, TOC parser, album-DR rounding, block lengths, live MusicBrainz lookup
tests/test_cd_path.py the whole CD path against a stub ripper: Linux layout, macOS layout (cd-paranoia only, no /dev/sr*, autosensed drive), explicit device, unparseable TOC, read-error exit status

tests/test_cd_path.py needs no optical drive — it fakes one — and its key assertion is an invariant rather than a hardcoded number: measuring a disc must give exactly the same DR values as measuring the same audio as plain files, which catches any bug in the rip/wiring path.

Not yet verified on real hardware: the CD path was developed where no optical drive existed (Linux container) and no Mac was available, so the cdparanoia invocation, the TOC parsing and the macOS autosense path have only been tested against the tools' documented output (including their source) plus a simulated drive in both layouts. If the first real run misbehaves, the useful diagnostics are drcd --verbose and cdparanoia -Q / cd-paranoia -Q run by hand.

Known limits

  • Pre-emphasis and HDCD flags on a disc are not applied (a pre-emphasised CD will measure a few tenths of a dB differently than a de-emphasised rip).
  • MP3/AAC sources measure the decoded audio; decoders differ by a hair, so lossy values can wobble by ±1 DR versus somebody else's rip of the same file.
  • The last sample of a file is ignored in the block analysis — inherited from the reference implementation on purpose (it changes nothing measurable).
  • Measuring takes roughly 1–2 minutes per CD (rip time dominates on slow drives; analysis is ~8 s for a full disc on a modern CPU).

License

GPL-3.0-or-later — see LICENSE. The measurement procedure is the one documented by the Pleasurize Music Foundation and implemented by the GPL-licensed dr14_t.meter and simple-dr-meter projects, so this implementation is released under the same terms.