i dont understand how peaking and clipping happens

[cns00]

i am hearing that the loudness should not be more than 0db. for example if there is a movie file and you do volume analysis on it and you get volume max to be -2db then you can increase the loudness by only 2db because -2 + 2 is 0. however, an increase of 2db is barely noticeable

please see https://www.youtube.com/watch?v=xHBY6q0HUyE. the guy said "to make the audio louder in audacity you can add gain but i dont recommend that because the sound will go over the peak or 0db and it will sound distorted"

i don't get that. if add a big amount of gain like for example +100db then the audio will sound distorted and squeaky. is it ok if i add a small gain like +15db? this is over the peak but it should still sound ok

[davexnet]

You don't want the audio to clip, but it can be amplified in a non-linear way, for example,
the compressor in Audacity, with the make up gain activated

[jagabo]

is it ok if i add a small gain like +15db? this is over the peak but it should still sound ok

It's up to you. And it depends on the nature of the recording. If only a small portion of the audio track is blown out you may find it acceptable (you may not even care about that portion -- say a click from a vinyl recording). If a large portion is blown out it will be much more audible. And how good is your audio system and hearing? If you're a golden ear listing on a decent system it will be more noticeable than if you have bad hearing and crappy speakers.

Here's a short sample that was normalized to 0 dB. And then another 15 dB of gain added. Can you hear the distortion? Do you find that acceptable?

[cns00]

Can you hear the distortion?

i heard it. it's obvious

[davexnet]

Can you hear the distortion?

i heard it. it's obvious

Then don't allow your audio to clip - it's as simple as that

[Cornucopia]

In the digital world 0dB is the ceiling, the upper limit. (A sample point at zero has all of its values at F, or FF, or FFFF; there is no value higher that can be described.)

If had a watermelon and wanted to have it stuck to the ceiling I can use a hydraulic lift to "push" it up there. It barely touches the ceiling. Done!
But if I am dissatisfied and want the bulk of the watermelon to be higher than it is (because it currently still has much of its mass below the ceiling), I can push the watermelon higher. What happens? The ceiling doesn't move, the watermelon starts to deform and then get crushed as it continues to move higher. Deformation might be acceptable to some and not with others, crushing is almost always unacceptable to all.

How high do you want to push your watermelon? Or can you just be satisfied with it being on (just touching) the ceiling?
"But I only want to push it 6 inches higher!?..."

If the analogy wasn't clear: the signal is the watermelon. Normalizing gets it to the ceiling. Compressors, etc deform, possibly to the point of crushing, possibly not (Zucchini shaped watermelon anyone?). Simple additional gain will quickly start crushing (clipping) the signal after even minor deformation has started.


Scott

[pandy]

Not sure what do you mean by peaking - based on description and thread discussion i may assume this is about ringing - if yes then perhaps this may help you understand root cause:

https://en.wikipedia.org/wiki/Ringing_artifacts
https://en.wikipedia.org/wiki/Transient_response
https://en.wikipedia.org/wiki/Impulse_response
https://en.wikipedia.org/wiki/Step_response

Clipping is reaction for insufficient system dynamics and it will occurs in analog and digital domain.

In digital domain maximum level that can be expressed without clipping is 0 dBFS - why it happens is explained in articles bellow:

https://en.wikipedia.org/wiki/DBFS
https://www.audioholics.com/audio-technologies/the-case-for-not-going-above-0-dbfs-for...ayback-systems
https://www.audioholics.com/audio-technologies/issues-with-0dbfs-levels-on-digital-aud...ayback-systems
https://service-tcgroup.tcelectronic.com/media/Level_paper_AES109(1).pdf

To simplify - if you wish to avoid any issues with sampled audio, normalize your samples up to max -3dBFS (efficiently you will loose half bit of the overall system resolution), in presence of the noiseshaped audio you can also choose to normalize your samples at -6dBFS (you loosing 1 bit of system resolution but you making place for noiseshaping signal preventing overall system clipping).

[jagabo]

With signed 16 bit sampling the smallest value that can be represented is -32768, the highest is 32767. To prevent clipping all samples must be between those extremes. When you amplify the audio the existing numbers are multiplied. If the result falls outside that range it will be "clipped". Any value over 32767 will become 32767. Any value below -32768 will become -32768. If you zoom into the a clipped waveform you will see flat tops on all the peaks, flat bottoms on all the valleys.


[Attachment 66583 - Click to enlarge]

Those flat tops create harsh sounding overtones, distortion.

[pandy]

With signed 16 bit sampling the smallest value that can be represented is -32768, the highest is 32767. To prevent clipping all samples must be between those extremes. When you amplify the audio the existing numbers are multiplied. If the result falls outside that range it will be "clipped". Any value over 32767 will become 32767. Any value below -32768 will become -32768. If you zoom into the a clipped waveform you will see flat tops on all the peaks, flat bottoms on all the valleys.


[Attachment 66583 - Click to enlarge]

Those flat tops create harsh sounding overtones, distortion.

This is more complicated - your case is typical hard clipping i.e. signal is beyond quantization range, but signal can be within quantization range and it will be clipped anyway.