Clipping vs True Peak: A Whistle Recording Experiment

One recording, three controlled versions

The source is a 12.98-second phone recording of a person whistling. It arrived as mono AAC in an M4A container at 44.1 kHz. I decoded it once to PCM, then made two deliberately simple variants: one raised by 0.75 dB, and one raised by 6 dB and written to 16-bit PCM so peaks above full scale were hard-clipped.

This is not presented as three independent recordings. The second and third files are controlled derivatives of the same performance, which makes the peak comparison easier to interpret. No limiter, compressor, denoiser, EQ, or normalization was used.

The measured difference

The original peaks at −1.27 dBFS. Raising it by 0.75 dB moves the peak to −0.52 dBFS, but still creates no clipped samples. The +6 dB copy reaches 0.00 dBFS, contains 468 clipped samples, and groups those samples into six separated events.

In this particular source the reconstructed True Peak estimate is effectively the same as sample peak for the first two files. That is a valid result—not every signal produces a large inter-sample overshoot. The hard-clipped copy reaches +0.08 dBTP in the current estimate because reconstruction between its flattened samples can rise slightly above digital full scale.

What each number actually tells us

Sample peak answers whether a stored sample reaches digital full scale. True Peak estimates what the reconstructed waveform may do between samples. A clipping detector looks for samples at or extremely near the ceiling, then groups neighbouring samples into useful events. These are related checks, but they do not answer the same question.

The timestamp list is operationally useful: clicking 5.23 seconds should take an editor to the first affected passage rather than forcing a search through the entire file. The count of 468 samples is evidence of clipping severity, while the six time groups describe where the damage occurs.

How to repeat the experiment

Keep the same decoder and analysis settings for all three files. Compare the files at matched monitor loudness if you are judging sound quality; louder playback alone can make the gained versions seem subjectively better.

1. 01

   Measure the decoded original

   Load the decoded source and record its sample peak, reconstructed True Peak, and clipping count.

   

2. 02

   Repeat with the near-ceiling copy

   Analyze the +0.75 dB version with unchanged settings and verify that headroom falls without producing a clipping event.

   

3. 03

   Open the deliberately clipped copy

   Load the +6 dB hard-clipped file and open its timestamped findings.

   

4. 04

   Verify each marked event

   Click a finding, listen around its marker, and inspect the corresponding spectrogram passage instead of trusting the label alone.

   

5. 05

   Export only after review

   Confirm the source identity and selected assessment profile, then export the report and supporting data.

   

Method limits

The source was recorded at 44.1 kHz and is not a standardized conformance stimulus. True Peak is therefore reported as the analyzer's reconstructed engineering estimate, not as certified broadcast-meter performance. AAC decoding and the phone's original processing are part of the source history.

Hard clipping is easy to demonstrate, but audible damage depends on duration, repetition, programme content, and playback level. A zero clipped-sample count also does not guarantee that clipping never occurred earlier in an analogue stage or during a previous encode.