Audio Recording

A clipping is a signal about 0dBFS that results to digital distortion or what simply called as a “clipped signal”. Why they sound so bad? What will happen to the original recording if it will become clipped? This post will examine the reasons behind by doing a simple experiment. I will use Adobe Audition software to perform this test and use a sine wave test tone as the model.

The Experiment: Using Sine Wave Test Tone

Step1.) Launch Adobe Audition. Go to File – New. Select 32-bit float/96KHz mono type for recording.

Step2.) Go to Generate – Tones. Select “Sine” as the flavor of the wave with 440Hz as its tone. Then set the peak amplitude of the sine wave to around -6dBFS, so it won’t clip. And set the wave duration to 5 seconds only, the rest will be at default settings. Press OK to implement the settings and also save the waveform as original.wav in a 32-bit float Windows WAV format by going to File – Save As .This is the resulting waveform:

Step3.) Close the file and re-open the 32-bit float wav file. If you play the sine wave tone it will only have one dominant frequency and that is the 440Hz steady tone. This is confirmed by looking at the audio frequency spectrum of the wave below. You will see that the most energy is concentrated in 440Hz (see the sharp peak at 440Hz). The rest of the frequencies are not significant to the resulting tone.

Step4.) Supposing you intend to clip this wave by applying a 10dB boost on the entire 5 second duration then saving it as clip.wav in 32-bit float WAV format. Close the file & re-open clip.wav in Adobe Audition. Below is how the waveform looks like in close-up view. As you have observed, the sine wave now looks similar to a “square wave”. The peaks of the sine wave are gone; it is “clipped-off” because it exceeds the maximum limit of 0dBFS.

Do you mean a clipped audio sine wave becomes a square wave? Yes but bear in mind that if the clipped audio wave is similar to a square wave, it possesses square wave frequency characteristics below (done by generating a square wave flavor instead of a sine wave, simply follow Step2). You will see a lot of harmonics in the frequency which is completely different to a sine wave frequency characteristic.

Step5.) Let’s check if the clipped 32-bit float sine wave now possesses square wave frequency characteristics by doing another frequency analysis. The answer is NO (look the frequency characteristic below). You will see that it still follows the sine wave frequency characteristics like in Step3. There is no significant alternation in frequency spectrum of the clipped sine wave. If you listen to this wave, it still sounds the same only that it becomes so loud. Does this mean that clipping does not result to digital distortion? It’s not yet conclusive, go to step 6 and step7.

Step6.) Let’s do a sample rate conversion and dithering to 16-bit/44.1KHz because this is the target bit depth and sample rate for all music production projects at the end of mastering. You can use a high quality sample rate converter such as Voxengo R8brain. Rename the resulting result as dithered.wav, now in 16-bit/44.1KHz wav format.

Step7.) Open dithered.wav file in Adobe Audition and do a frequency analysis again. Guess what? The clipped sine wave test tone at 16-bit/44.1KHz NOW possesses a square wave frequency characteristic; see below. In short, your original sine wave test tone is NOW DISTORTED. In real audio and real music production, the generated harmonics of the square wave actually is very unpleasant to listen and that is the cause of undesirable distortion you will hear when an audio wave is clipped.

Lessons Learned in Digital Audio Clipping

Fans of loudness wars and engineers that would like to push the peak amplitude beyond 0dBFS simply do not know the consequences. The truth is that at high resolution like a 32-bit float, you won’t notice this distortion because of the very big dynamic range associated with high bit depth (32-bit float). It simply sounds loud without distortion which is good. But most DAW like Apple logic can deal with signals above 0dBFS without distortion PROVIDED the bit depth is floating point (32-bit float). This is why distortion is not caught and audible at higher bit depths when you are listening to it or viewing the frequency spectrum.

However the truth would be revealed when you down sample and dithered the clipped audio down to 16-bit/44.1KHz. It would result to distortion which you can hear in your studio monitors and consumer speakers degrading audio quality. In the above experiment, it has been shown that a clipped audio will possess “square wave like” characteristics that significantly alter the original sound of the recording (in other words, it becomes distorted). Simple rules you can apply is setting headroom:

1.) When tracking, allow some -6dBFS headroom (max peak set to -6dBFS).
2.) In mix down, set the maximum peak to -3dBFS.
3.) In mastering set the maximum peak to -1dBFS at most. Some engineers are happy around -0.3dBFS but never at 0dBFS or beyond because of higher chances it will clipped during down sampling, dithering and MP3 conversion. I will write a separate tutorial about the optimum mastering peak amplitude that won’t result to distortion in both WAV and MP3 audio formats.