Audio silence on 1.0

I have audio files with silence on 1.0 (max amplitude). I attach an example of audio waveform. I don't know why it is on 1.0? How can I convert it to have silence on 0 (as usual)? I noted that at the beginning of the waveform it is starting from 0 and goes to 1.0 - any ideas why? The files are telephone recordings, they are sampled with 8kHz and coded with GSM. I don't have more information about recording procedure.

In addition to Olli's suggestion, it might also be a by-product of the encoder's processing.

Removing this offset would ammount to a simple removal of the DC component.

This is practically done with a high pass filter with a cut-off (or rather pass-through) near zero Hertz.

If you are working in Audacity, guessing by your screeshot, this is done by Effect->Normalise and from the window that pops up, make sure that you check "Remove DC Offset".

In terms of digital signal processing, you can calculate the arithmetic mean of the whole sample and then subtract it from all values.

(Please note: This is assuming a constant DC offset across the recording. If you also happen to have DC drift, it would be better to perform a "best line fit" to the whole sample and then subtract that estimation from the whole sample to "align" it back to zero.)

Hope this helps.

For transmission of audio by amplitude modulation of for example light, the audio can be made unipolar in order to make silence correspond to no transmission which for some receivers means no noise during silence. Just a guess but your audio may have gone through such processing. From the look of the waveform it may have worked like this:

Start with bias = 0. For each time step:

• bias = bias + constant, where constant determines the slope seen during silent segments in your graph
• If bias + input > 1, then set bias = 1 - input to avoid clipping.

It is not possible to exactly recover the original audio from the output of such processing, but you could use a similar method that keeps shifting the bias towards the direction that centers the audio, and changes the bias appropriately any time the audio would be clipped.