4
$\begingroup$

Hi I'm doing a past exam question about converting from stereo to mono. The scenario is below:

enter image description here

The first question is: Explain why would this result in a noisy encoding. How is the noise introduced and what does it sound like?

Second Question: What would you do to prevent such noise being introduced?

Answer 1:

The only thing I could think of on this is that the two channels could be slightly out of phase and therefore interfere destructively when summed. I'm not sure what this would sound like but I guess the amplitude and frequency are being changed by this so I know it wouldn't sound right!

Answer 2:

We would have to do something to remove the phase difference, right?

$\endgroup$
  • $\begingroup$ I wouldn't skip every other sample. Rather, average successive samples. As to the left-right phase issue, that's above my pay grade. $\endgroup$ – Daniel R Hicks May 27 '12 at 13:20
  • $\begingroup$ thanks for the answer - but why is this preferable to skipping every other? $\endgroup$ – user1058210 May 27 '12 at 13:25
  • 2
    $\begingroup$ If you skip samples you create the aliasing situation described by pichenettes, plus you effectively amplify any noise in the signal (which is sort of the same thing as aliasing and sort of different). $\endgroup$ – Daniel R Hicks May 27 '12 at 14:17
  • $\begingroup$ There's nothing wrong with averaging the two channels together. Maybe rename this question? $\endgroup$ – endolith Jun 1 '12 at 0:40
21
$\begingroup$

There are indeed rare instances of recordings in which the L&R channels are in opposite phase. This was done in the late 70s and early 80s to create a cheap "spatial sound" effect (like the "surround" button on low-end hi-fi systems of the time), and on such recordings, computing the average of the left and right channels gives a null output. Some stereo effects (like chorus or reverb) would also make a recording sound less lively or flatter after a mono conversion, but this doesn't really qualify as noise. All these are rare problems, and averaging the L&R channels is the standard way of converting from stereo to mono. In the past I've worked on a system that handled large catalogues of audio files, and my approach was to check whether the energy of the average of the channels was below a certain fraction of the energy of the individual channels ; and if so, used the difference instead of the sum as a mono conversion. Ideally you could compute the cross-correlation between the left and right channels to find the lag at which they are maximally correlated, and delay one channel by the lag before the mixing so they are maximally in-phase with each other, but this is overkill for just handling the 15 or so tracks out of 100,000 that had a massive phase inversion problem - to give you an idea of the rarity of the problem...

The most obvious error is converting from 44kHz to 22kHz by skipping every second sample. This will alias all frequencies in the 11kHz - 22kHz range, and will make the downsampled recording sound "metallic", "crunchy" or "gritty". A proper way of downsampling a recording is to first apply a low-pass filter to remove all frequencies above half the target sample rate ; then decimate (skip every second sample).

$\endgroup$
  • $\begingroup$ That was a really interesting answer - thanks! Also, I can't believe I missed the aliasing...! $\endgroup$ – user1058210 May 27 '12 at 13:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.