Measure a recording Method
How the measuring works
The tool runs the WPAudio Engine on your device. This is what it measures, what it is checked against, how closely it agrees, and what a report can and cannot tell you.
01
What it measures
Every report holds the same 38 measurements. Loudness, loudness range and true peak follow ITU-R BS.1770 and EBU R 128. The rest describe the file itself: its peaks and average level, its silences, any samples at full scale or invalid samples, the balance and width of the stereo picture, and how its energy spreads across the spectrum. Where a measurement follows no standard, the report says how it is defined.
A sample at full scale, the highest or lowest value the file holds, is not clipping by itself: a peak brought to exactly full scale reaches it at one sample, or at two. A report calls it clipping only when 3 or more samples in a row, in one channel, hold the same value at full scale, and says where the first and the longest run are. Above a sixth of the sample rate a wave has fewer than three samples per half cycle, so clipping there shows only as samples at full scale and a high true peak.
From the measurements come the findings: a short list of what is worth checking, the most serious first, each with what it is, why it matters and what to do. The words are written for musicians first, and every technical term they use has its plain meaning in the engine's glossary.
02
What it is checked against
The loudness meter is checked against the 100 expected values of the EBU's test signals, EBU Tech 3341 and EBU Tech 3342, and every one is inside its tolerance. On the signals that state a value, the engine holds its integrated loudness to 0.03 LU and its true peak to 0.06 dB of the expected value, closer than the documents ask.
Where the engine agrees with the Studio's earlier implementation, each measurement has a tolerance: how far it may stray from that implementation's value on the engine's test files. Three more tolerances hold the engine to the EBU test signals. All 27 are at or below one displayed unit, the last digit the report shows.
Where the Studio's earlier implementation did not follow the standard, or was not sound, the engine departs from it on purpose: 14 measurements, each by a written decision with the differences measured. For example, a file shorter than the measurement window gets no loudness value rather than one padded with silence.
03
The same answer on every device
The engine that runs in your browser and the engine's command-line build give byte-identical reports on all 34 of the engine's test files: every container and bit depth it accepts, every sample rate it is tested at, and the files it refuses.
On the laptop it was measured on, the engine read a ten-minute stereo file at 50 times real time in Chromium, the browser engine of Chrome. The engine's own memory grows with the sample rate, and hardly with the length: it was 13.2 MB at the end of the longest and densest file it takes, three hours of 24-bit stereo at 192 kHz, measured in Chromium on the same laptop, and 4.3 MB after a sixty-minute file at 48 kHz, measured in Node. In Chromium, Firefox and Safari's engine, Cancel handed control back within 0.8 to 10 milliseconds. Phones have not been measured yet.
04
What a report shows, and what it does not
A report carries the SHA-256 of your file's bytes and every value the engine measured. Anyone can measure the file again and see whether the report is true of it, and whether it was changed after it was made. A report does not show who made it: nothing is signed, and anyone can make a fresh report of any file.
In a report's data, every measured value is written as decimal text, such as "-14.2", not as a number (the one yes-or-no measurement is written as yes or no). The text is the value exactly as the report shows it, at its stated precision. That is what makes a saved report checkable byte for byte: the same engine build measures the file again, writes the same text, and compares the two. A number would be read and written again by whatever program opened it, which can change its last digits or its form.
A report cannot tell whether the music is good. Where it compares a recording with a streaming service, it says what the service says it does, with the date that was read; services change how they play music.
Measure a recording · The open-source code inside the engine