ETM Learn · 07 / 08
Separate clipping, moving gain, sample peaks, reconstructed peaks and loudness.
After compression, examine control of peaks relative to a ceiling. A limiter usually moves gain over time. A clipper instead limits amplitude values themselves. Both can alter sound, so listen for the cost as well as watching the meter.
Digital audio is a sequence of samples, values at separate instants. Sample peak is their largest absolute value. The reconstructed wave may peak between samples; estimates that maximum. Start with the correct limiter mode and raise Drive gradually. Then explore Lookahead, which delays audio relative to detection.
Dotted terms offer a definition on hover, keyboard focus or tap. Click to keep it open; Escape or an outside click closes it. All definitions are also available here.
An estimate of the reconstructed waveform's peak, including between samples. Accuracy depends on reconstruction; this teaching meter is not certified.
Peak-to-RMS ratio over the same measurement interval. Describes the peak relative to effective level, not perceived loudness alone.
Loudness units relative to full scale using weighting and time aggregation. Specify whether the measurement is momentary, short-term or integrated.
Select limiter and the correct kernel. Raise drive gradually and compare waveform, gain reduction and spectrum.
Predict → change one control → compare → explain the result.
Intentionally faulty modes violate conditions in this implementation. They do not describe every limiter. Reconstructed peak and loudness readings are estimates.
More lookahead permits attenuation before the delayed peak, but affects a longer neighbourhood and costs latency. This is pre-attenuation, not filter pre-ringing.
Set 0.25 × fs and 45°. The reconstructed sine can exceed its largest sample. This is ideal sine reconstruction; actual filters and lossy coding can change the result.
Does sample peak below 0 dBFS guarantee reconstructed peak below it?
Extrema of a band-limited waveform need not occur at sample instants.A nonnegative normalized smoothing kernel confined to a suitable window can preserve the ceiling in this running-minimum design. These are sufficient conditions here, not necessary conditions for every limiter. IIR smoothing is not inherently oscillatory, but smoothing without a suitable safety constraint can miss peaks.
The peak meter uses finite 8× windowed-sinc interpolation with periodic extension. It is not a certified BS.1770 meter: finite filtering and time resolution limit accuracy. Loudness uses mono K-weighting and gated 400 ms blocks; loop length and startup affect it. This is not a full stereo programme measurement.
is 20 log₁₀(peak/RMS), with consistent peak definition and interval. PLR is dBTP minus integrated , a different metric. Normalization depends on service, mode and device. No universal streaming target or ceiling exists; check the current delivery specification and listen to the processing.
CF = 20 log₁₀(peak / RMS)
Divide peak amplitude by RMS of the same interval and convert to dB. A steady sine has ratio √2, about 3.01 dB. For total silence the ratio is undefined.
PLR = peak[dBTP] − loudness[LUFS integrated]
Subtract integrated LUFS from peak dBTP. Example: peak −1 and loudness −13 give 12 dB. This differs from crest factor because LUFS is not plain RMS.
Before scaling, the largest sample is 1/√2 of the sine amplitude.
Scaling is √2, so the reconstructed peak is √2 and 20 log₁₀(√2) = 3.0103 dB. This is an ideal band-limited sine calculation, not a universal inter-sample overshoot bound for music.
Subtract RMS level from peak level, using consistent amplitude references.
−1 − (−13) = 12 dB, corresponding to peak/RMS = 10^(12/20) ≈ 3.981. Replacing RMS with integrated LUFS produces a different quantity, PLR, and cannot be justified by changing the label alone.
Further reading: ITU-R BS.1770