I've been using Sox v.14.3.2 to de-emphasise CDs that have the emphasis flag set. After a couple of dozen or so such discs, I reckon that the resulting music file sounds rather flat - i.e. with reduced top end - compared to other CDs where the preemphasis flag is not set. Am I alone in thinking this? My implementation is Sox inputfile.flac outputfile.flac deemph. What do others think?
I've yet to run into any CDs with the de-emphasis flag set, was not very commonly used except on a few early CD classical recordings.
I guess I am wondering why you are manually applying de-emphasis to those flac files? Most dacs afaik still do this automatically if the flag is set and it doesn't need to be done twice.
Perhaps I've missed something, can you elaborate on why you are doing this? I've been ripping to flac via EAC or J River for about 7yrs now and never used sox or any other tool to manually de-emphasize anything.
(Also have used sox as a player in linux...)
I guess I am wondering why you are manually applying de-emphasis to those flac files? Most dacs afaik still do this automatically if the flag is set and it doesn't need to be done twice.
Perhaps I've missed something, can you elaborate on why you are doing this? I've been ripping to flac via EAC or J River for about 7yrs now and never used sox or any other tool to manually de-emphasize anything.
(Also have used sox as a player in linux...)
The biquad filter sox uses is working correctly if you don´t use some very old buggy version.
From the sox page: "The current deemph effect works fine and is accurate to 0.06dB"
This is correct. Depending on the reference curve used sox is always correct. Beware of the iTunes implementation. In some version the de-emphasing dips ~1dB to low.
I have done some graphs a while back and posted in other forums for this one iTunes version. I don´t know about later versions.
After all it does matter more what way was used to apply pre-emphasis. It seems every curve with sample sinuids i collected over time for testing was way more off as the sox de-emphasis itself.
In the end i decided to apply a de-empasis that is more like +0,10 dB to be on the safe side. Better a tiny bit to much highs as to less.
From the sox page: "The current deemph effect works fine and is accurate to 0.06dB"
This is correct. Depending on the reference curve used sox is always correct. Beware of the iTunes implementation. In some version the de-emphasing dips ~1dB to low.
I have done some graphs a while back and posted in other forums for this one iTunes version. I don´t know about later versions.
After all it does matter more what way was used to apply pre-emphasis. It seems every curve with sample sinuids i collected over time for testing was way more off as the sox de-emphasis itself.
In the end i decided to apply a de-empasis that is more like +0,10 dB to be on the safe side. Better a tiny bit to much highs as to less.
PC playback doesn´t know about a flag and applying de-emphasis until you have an application that reads the "FLAGS PRE" info in a CUE file. So the played files will be most likely played back horrible sounding.
The dacs have the ability, but have to be told by the driver/playback software to deemph.
I was under the impression that FLAC files have no facility for storing pre-emphasis information. If I don't use Sox preemph, it usually sounds too bright, so I don't think it's being applied twice.
I wonder whether some recordings actually used proper pre-emphasis at all or just sounded a bit bright so the engineer set the preemph flag? (eg. Cala's disc of Franck's Symphonic Variations, rec. 1987)
In a couple of cases I wonder if the flag was set by mistake (eg. BBC Music magazine's CD of Stravinsky's Firebird Suite, rec. 2006)
Surely someone else has noticed this sort of thing?
I don´t have many cds with pre-emphasis but the ones that were detected by EAC were absolutely right.
I think it is more the other way around and many people play back files that never were checked for it.
Most likely on many so called Audiophile systems it shouldn´t matter, the system itself will color the sound more as the de-emphasis.
I think it is more the other way around and many people play back files that never were checked for it.
Most likely on many so called Audiophile systems it shouldn´t matter, the system itself will color the sound more as the de-emphasis.
This misses the point that if de-emphasis is applied incorrectly, these recordings will acquire a different tonal balance from the rest of the music collection.
My observations are that without de-emphasis, these recordings sound brighter than the norm, but with sox de-emphasis they end up sounding noticeably duller than the norm. Why should this be?
This shouldn´t be. When i listen the non-de-emphasis version directly before the de-emphasised version i also feel it cuts to much highs. That is normal. We like highs and will prefer it. Also you may prefer the unmodified simply because it seems louder!? When i am at it i also normalize my resutling files.
If i listen the recording without direct comparison on my system the mofified files sound right to me with the ones i have.
Edit: maybe you want to share a test sample of 20 seconds, either with a music part you hear to be totaly wrong or simply a silent part with obvious noisefloor. The shape of the noisefloor if no straight line can hint to a degree but this needs some fiddling and luck to check.
Last edited:
This depends on local law i suppose. So far i sometimes uploaded it to the uploads section of of another forum i am in but any filehoster will do. Just cut a piece of the original, non-modified one. I hope it is long and obvious enough to judge.
View attachment raw excerpt.zip
Attached is a zipped 20 second snatch of a work for 2 pianos. EAC marks all tracks as pre-emphasised. This section is from the raw file produced by EAC, i.e. de-emphasis has not yet been applied. It sounds slightly tinny, nothing more. If you de-emphasise it with Sox, it sounds dead flat. Tell me what you think.
Attached is a zipped 20 second snatch of a work for 2 pianos. EAC marks all tracks as pre-emphasised. This section is from the raw file produced by EAC, i.e. de-emphasis has not yet been applied. It sounds slightly tinny, nothing more. If you de-emphasise it with Sox, it sounds dead flat. Tell me what you think.
Only crappy laptop speakers here so can't listen properly. However here's the spectrum before and after sox deemph—it looks as expected, gently rolling off to about 10dB:
So if it sounds flat, perhaps that's what was intended.
An externally hosted image should be here but it was not working when we last tested it.
So if it sounds flat, perhaps that's what was intended.
I am with highpass and pre-emphasis is applied to this.
For this piano piece the frequencies touched by the emphasis process are not that important and the sound doesn´t change as obvious as it would be on other instruments.
On the short very silent part from second 9.5 on the noisefloor rises obviously pretty steep up to roughly +10dB at 20kHz (look from 5kHz on)
When you apply de-emphasis this is a flat line as you´d expect.
For this piano piece the frequencies touched by the emphasis process are not that important and the sound doesn´t change as obvious as it would be on other instruments.
On the short very silent part from second 9.5 on the noisefloor rises obviously pretty steep up to roughly +10dB at 20kHz (look from 5kHz on)
When you apply de-emphasis this is a flat line as you´d expect.
I tried to make visible what i talk about.
Here is a pic of the very last 177ms of your sample. Unfortunately it isn´t completely silent. Some energy of the piano is still there so below 10kHz it may not be really conclusive. You may see it much clearer on a realtime FFT
If we only look at the noise in the range from 10kHz-20kHz you may see how the de-emphasis works for this piece of music.
Judging the sound alone i find it really hard to judge this solo piano with this short sample. Both versions don´t sound wrong in this case. There are countless recordings that sound much more off in either direction. More bright or more dull. Since it seems to be a valid pre-emphasis CD you should live with the music as it was inteneded by the label but that is only my opinion.
Here is a pic of the very last 177ms of your sample. Unfortunately it isn´t completely silent. Some energy of the piano is still there so below 10kHz it may not be really conclusive. You may see it much clearer on a realtime FFT
If we only look at the noise in the range from 10kHz-20kHz you may see how the de-emphasis works for this piece of music.
Judging the sound alone i find it really hard to judge this solo piano with this short sample. Both versions don´t sound wrong in this case. There are countless recordings that sound much more off in either direction. More bright or more dull. Since it seems to be a valid pre-emphasis CD you should live with the music as it was inteneded by the label but that is only my opinion.
Attachments
Last edited:
Good plot—mine was done over the whole sample.
Maybe it's just that the mic'ing wasn't so great on the original recoding, or that tastes have changed since.
Here's a thought: deemph is a low-pass shelving filter, exactly equivalent to 'treble -9.477 5283 0.4845s' (you can use the --plot option to check that the filter coefs are identical), so you could use 'treble -5 5283 0.4845s' for a 'halfway house' solution (or whatever peak attenuation value sounds best in this case).
Maybe it's just that the mic'ing wasn't so great on the original recoding, or that tastes have changed since.
Here's a thought: deemph is a low-pass shelving filter, exactly equivalent to 'treble -9.477 5283 0.4845s' (you can use the --plot option to check that the filter coefs are identical), so you could use 'treble -5 5283 0.4845s' for a 'halfway house' solution (or whatever peak attenuation value sounds best in this case).
Thank you very much for looking at this. The background noise analysis seems conclusive and I am really glad you suggested this. But I must try the same thing on a non-pre-emphasised CD to be sure that a flat trace is indeed expected.
Generally speaking I find that using filters leaves me feeling as though my ears have been twisted. I would only consider doing this where there is a very obvious error.
I have 39 CDs in my collection which have pre-emphasis, which is enough to form a conclusive, if subjective, opinion. Most of them, to my ears, demonstrate a very slight smoothness or dullness, compared to other CDs. It is not a matter of volume, as I always adjust that for the individual recording. Maybe it is the intended result of pre-emphasis, which I presume is a reduction in digital artefacts at the top end.
Just out of interest, what do you think that massive spike at 17300Hz is all about? Could that have anything to do with it? It is way above my hearing limit so obviously I can't hear anything, but could it be possible that it has an effect on my hearing?
Generally speaking I find that using filters leaves me feeling as though my ears have been twisted. I would only consider doing this where there is a very obvious error.
I have 39 CDs in my collection which have pre-emphasis, which is enough to form a conclusive, if subjective, opinion. Most of them, to my ears, demonstrate a very slight smoothness or dullness, compared to other CDs. It is not a matter of volume, as I always adjust that for the individual recording. Maybe it is the intended result of pre-emphasis, which I presume is a reduction in digital artefacts at the top end.
Just out of interest, what do you think that massive spike at 17300Hz is all about? Could that have anything to do with it? It is way above my hearing limit so obviously I can't hear anything, but could it be possible that it has an effect on my hearing?
Here it is:
Interesting that it's much stronger on the left channel and seems to dip in volume on the right channel. No idea what it is, a fault somewhere in the recording chain probably.
Unlikely to have any effect on what you hear: ultra-sonic is ultra-sonic, providing that your playback system is 'linear', which all strive to be.
You could easily notch it out (sox equalizer effect) if you're worried, but hey, another filter...
An externally hosted image should be here but it was not working when we last tested it.
Interesting that it's much stronger on the left channel and seems to dip in volume on the right channel. No idea what it is, a fault somewhere in the recording chain probably.
Unlikely to have any effect on what you hear: ultra-sonic is ultra-sonic, providing that your playback system is 'linear', which all strive to be.
You could easily notch it out (sox equalizer effect) if you're worried, but hey, another filter...
Noisefloors can have every kind of shape especialy since there are out tons of noiseshaped dither so this won´t work with every recording.
On the pre-emphasis cds i checked there was no sign of these dither curves. With your recording it is also unlikely because the hard cut-off at 20kHz. Dither wouldn´t stop there.
Since you reported the pre-emphasis is reported while ripping + the fitting of the noiseshape we must conclude the CD is right.
I ripped many CDs and on none i had a false positive for this flag.
The 17kHz spike is nothing special and in many recordings you see something like this. Most often it is some braun-tube working in a monitor or TV, in this case placed to the left. 17kHz maybe because there were 34kHz monitors around and this is one half-tone!? This would be a case we had some use form a HiRes transfer
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.