Methodology
This database measures original audio files bundled with operating systems and applications without gain changes or normalization. The source audio itself is not published.
Loudness
Integrated, 400 ms momentary, 3 s short-term, loudness range, and true peak measurements use FFmpeg’s ebur128 filter following ITU-R BS.1770-5 and EBU Mode. A silent tail is appended so a meter can settle after a short event; original samples are not altered. Short-term and LRA are unavailable for sources shorter than three seconds.
RMS, active segment, and level envelope
Full-file RMS is the root mean square across every sample and channel. Activity uses non-overlapping 10 ms frames and the higher of −60 dBFS or 40 dB below the peak frame RMS. Active RMS covers the interval from the first through the last active frame. The time plot uses non-overlapping, channel-combined short-window RMS. Frames are at least 10 ms and grow automatically for longer sources to stay at or below 400 points. The chart uses a fixed 0 to −80 dBFS display range for comparison, while JSON retains values below the display floor.
Spectral characteristics
A Welch power spectral density estimate uses a Hann window of at most 4096 samples after arithmetic channel averaging. The database records broad-band and third-octave energy, spectral centroid, peak frequency, and the proportion of energy from 200 Hz to 8 kHz.
Limitations
These are measurements of bundled source files, not acoustic playback levels. Actual level relationships depend on OS, app, player, and output-device settings.
Values are not recommendations, safety judgments, or product rankings. Unavailable measurements remain null in JSON rather than being replaced by zero.