Subject:Resample Increases Gain?

Posted by: DonM1

Date:5/18/2007 6:08:49 PM

All:

When using the resample process I have found that sliding the interpolation accuracy to it's higest setting will increase the gain of the resampled audio. Why is this the case? From Sony's documentation it seems the highest setting should relate to the fidelity of the sampled result - but why gain increase? Thanks in advance for any assistance..

-D

Subject:RE: Resample Increases Gain?

Reply by: timo75

Date:5/19/2007 4:16:39 AM

Did you hear the gain release, or did you just see it on the waveform/level meters?

If the file was really louder, than it is a bug. If you just see higher peaks in the waveform, or clipping, than this is quiet normal. When applying high order interpolation, so called intersample peaks can occur. As the interpolation algorithm 'guesses' sample values between two existing samples, it can happen that it assumes a peak between them that exceeds every other samplevalue in the whole audiofile! This results in a waveform with higher peaks than the original waveform.

You can avoid this by using a limtiter that provides intersample peak treatment... e.g. the izotope Limiter that comes with SoundForge9, or the MPL-1 from Kjaerhus.

Hope this helps!

Timo

Subject:RE: Resample Increases Gain?

Reply by: DonM1

Date:5/19/2007 5:52:28 AM

T:

Thanks for the reply.

Why would hearing the increase indicate a bug? I anticipated the possibility of intersample peaks as you've mentioned. But why would the higher interpolation setting be more susceptible to that possibility?

Clearly the "energy" in the file is substantially increased. A 10 second file of white noise with -1 dbfs peaks will 'rise' to +5.3dbfs peaks when performing SRC at the highest interpolation setting. Since that is more than 40% increase I suspect hearing it is not a problem unless I am totally mistaken about what the meters are reporting and what I am hearding.

If the interpolation setting is aimed for higher fidelity (per the documentation) I would suspect fidelity would equal accuracy - so if intersample peaks are the issue wouldn't that also so up at any interpolation setting? Also the interpolation setting has different gain results when down sampling from higher sample rates. For instance, a 20 second 96Khz file of white noise -1dbfs only revealed two clips exceeding DBFS (+2 in both cases) when converted to 44.1Khz at the highest interpolation. It seems the 48Khz to 44.1Khz suffers the most from this gain at the highest interpolation setting.

You're right about using limiting. I am just trying to figure out what is really happening to the audio at the various interpolation settings when doing SRC.

-D

Subject:RE: Resample Increases Gain?

Reply by: DonM1

Date:5/20/2007 5:51:11 AM

If anyone is interested in this 'problem' I have posted an extensive test with screen captures at the Sonar Forum, where this topic first came up.

See:

http://forum.cakewalk.com/tm.asp?m=1053129&mpage=1&key=

Any input is appreciated!

-D

Subject:RE: Resample Increases Gain?

Reply by: timo75

Date:5/20/2007 9:34:33 AM

Hi DonM1,

you are measuring peaks, that has nothing to do with energy and loudness. You should check the RMS level which corresponds to the loudness you hear (that's why I asked if you hear or just see the difference).

The higher the interpolation order, the more intersample peaks occur, this is a mathematical problem. With linear interpoIation (2nd order), no intersample peeaks occur, they start with cubic interpolation... arggh, makes no sense, I cannot explain this in english, sorry... :-(

Try to limit your file before you downsample it, with an intersample peak limiter, e.g. one of the above mentioned. If you see then still significant gain rise, there might be a bug.

Subject:RE: Resample Increases Gain?

Reply by: DonM1

Date:5/20/2007 1:23:27 PM

T:
Absolutely correct - I realize I the DBFS meters are measuring peaks. And you are right again that the engery in a file is much more related to RMS and the peak to average ratio than do the peaks. (see page 65 Bob Katz's Mastering Audio text for more details)

I am concerned about the mastering process specifcally dealing with peaks. I am concerned that Sony's resample algorithms are changing the energy in the file (especially 48k to 44.1k sample rate conversion) and I'd like to know exactly why, and even more specifically I don't want to have to use a limiter just for SRC.

Another way to ask this, does Waves or R8Brain increase the energy in the file during SRC?

-D