I am working on a research project and I was required to record a few seconds audio from a few volunteers in a studio. When I checked the recordings I realized the audio was clipped in important regions of the recording.
Is there any way to recover the data?
Also, can someone explain to me what difference clipping makes on the frequency spectrum of the audio and how reliable is the clipped data?
First of all, let's examine what happens when a signal clips: An audio signal can be represented by a string of numbers, so let's say you wan't to record this signal:
0 0.5 1.0 1.2 1.0 0.5 0
But the signal clips, so what you get is this:
0 0.5 0.9 0.9 0.9 0.5 0
See what's up? Suddenly no signal is present above 0.9...
The thing is, when a signal clips you loose information. A byproduct of this is added artifacts, like higher order harmonics (which introduce audible distortion).
When using de-clipping processing, you can successfully remove and/or mask the artifacts, which can make the signal more perceptually acceptable. But you are not recovering the missing data, only treating the byproduct.
So getting back to your questions:
Is there any way to recover the data.
Nope, unfortunately not.
Also, can someone explain to me what difference to the frequency spectrum the clipping will make?
As stated above (and by Andrés G. Duarte) you'll see added harmonics at higher frequencies.
How reliable is the clipped data?
It won't be a realiable representation of you input data. Whether it is usable, however, depends on the application. For research I would definitely redo the recordings. You are introducing unknown error sources by using clipped signals - and research have enough unknown error sources already :-)
I would suggest that if the audio data is to be used in a research project in which having not clipped audio data is of some importance, you should start again.
Although you might use some tools to make the clipped files more audible, you actually have no idea of what the data is in the clipped zones (unless that they are over the maximum input level).
Is there any way to recover the data.
Yes, you can use de-clipping software, e.g. Waves or Izotope plugins created for this purpose.
what difference to the frequency spectrum the clipping will make?
"As the signal simply "cuts" or "clips" at the maximum capacity of the amplifier, the signal is said to be "clipping". The extra signal which is beyond the capability of the amplifier is simply cut off, resulting in a sine wave becoming a distorted square-wave-type waveform."
"In the frequency domain, clipping produces harmonics at higher frequencies than the unclipped signal. This additional high frequency energy has the potential to damage a loudspeaker via overheating." https://en.wikipedia.org/wiki/Clipping_(audio)
How reliable is the clipped data?
Reliable for what? if you mean for recovering the un-clipped data, it depends on how much clipping you have.
Generally no. If the clipping was only one sample the CIA might be able to do it.
Clipping causes high order harmonics. Look at the FFT of a square wave versus a sine wave. The more you clip the more that sine wave looks like that bunch of squiggles from the square wave.
The isotope RX restoration software has some de-clipping features - though any sound coming in at a greater volume that the clipping point threshold is not actually stored and will be digitally synthesised - the best solution would be to redo the recording session.
