Some may remember I was asking a load of questions a while back trying to get Audacity set up for vinyl ripping.
Nearly 40gb later I can report that I am delighted with the quality; it is really very, very good indeed. But it is in further post processing I am getting poor results, using the same workflow I successfully used on my first ~150gb of rips 2001-2008. I wonder if anybody can help me to pin down where I am going wrong?
What I am doing is removing clicks in Audacity and saving 16bit .wav files ripped via an Asus Xonar D2 soundcard.
I am then importing these into a program called Steinberg Clean, that allows you to apply various noise removal filters, EQs etc, and save the output. I realise others use different tools for vinyl clean-up, and am always open to suggestions, but Clean suits me well as it is simple and effective and has always served me well.
What I am getting is heavily distorted bass, using approximately the same EQs I have always used.
I believe what is occurring is that my new Asus soundcard is much more sensitive and capturing lower frequencies compared to my old Soundblaster card, and that Clean is somehow distorting these into the audible spectrum, when I tweak the low frequency EQ (even applying negative gain appears to produce distortion).
I suspect the solution may be to apply an EQ cut off in Audacity (hopefully via a chain) but would like to quantify settings.
I have been on holiday but having researched my problem further had started to believe there were in any case too many variables to be able to pin down a solution; I really appreciate you trying to help me out.
To answer your question yes, on the most heavily affected tracks there is in fact some distortion prior to processing in Clean, which then accentuates this.
It also seems clear from having looked at a number of frequency distributions that the issue almost certainly arises in my Pro-ject pre amp instead of the Asus soundcard (I had previously been using a Vestax PMC 40).
Attached is a sample from one such track (I can also post an audio clip if that helps)
That’s one of the things I am a little confused about. For example, digging out the spec on my old but still very good Vestax PMC 40 mixer, I see it supposedly has a frequency response of 20-20,000Hz. But looking at the spectrum of .wav file recorded with this in Audacity I see there is plenty going on under 20Hz, also right down to sub 3Hz as per the spectrum using my current setup posted above. Where does this come (through?) from?
Furthermore, many of my old .wav files also have a hump in the spectrum around 12-15Hz, which I suspect is the area causing me problems, but these max out at around -48db. As with frequency is there rule of thumb regarding what level of db is audible?
FWIW I have been trying to trim this stuff with the Audacity EQ and have used two presets depending on the magnitude of that hump. I have also tried to experiment with the high pass filter but am having trouble configuring the parameters, especially the roll off. Do you have any suggestions?
And coming back full circle to the business about audio equipment frequency response under 20Hz, I nevertheless believe - admittedly subjectively - that many of my old vinyl rips sound better than CDs. I am therefore reluctant to cut off more than necessary of the low end frequencies, if that is the reason. . .
– Wikipedia –
IEC RIAA curve
An improved version of the replay curve (but not the recording curve) was proposed to the International Electrotechnical Commission with an extra high-pass filter at 20 Hz (7950 µs). The justification was that DC coupling was becoming more common, which meant that turntable rumble would become a greater problem. However, the proposal did not achieve traction, as manufacturers considered that turntables, arm and cartridge combinations should be of sufficient quality for the problem not to arise.
20 Hz is jet engines, earthquakes, or that one low organ note that you can’t hear so much as feel. Chances of those musical notes making it onto and off the vinyl record are zero. There were reports of people trying and producing vinyl records that nobody could track without putting a penny on the headshell. The stylus would literally leave the groove from the abuse.
So you probably have award-winning motor rumble and should get rid of it. As in the article, Audacity will cheerfully “edit” battery DC voltages and yes, it causes no end of problems. High pass filter at 30Hz and let it droop to the left, probably either 12dB or 24dB per octave.
Another real world note. You can very nearly not hear a 6dB change. It’s very gentle. That’s from 1 to 0.5 on the Audacity default timeline.
“DC” stories abound. When I lived back east (US), one of the engineers fascinated with his new Crown DC-300 power amplifier decided to do a real-world test. He connected a 1.5v penlight battery to the input of the dc-coupled amplifier and amplified it to 6v making it the most expensive lantern battery on earth.
Another is the thought experiment of what would happen if you actually had a sound system that would go down to DC. You wouldn’t need a fan or a vacuum because DC sound is wind. You could use your speakers to blow-dry your hair.
That gives you a colorful idea of what kind of trouble you can get into with a sound system that goes down too low.
Put that kind of audio into a powerful conventional sound system without warning everybody and you’ll be picking up the woofer cones from the other side of the room – and you won’t be able to hear for a while.
Thanks very much Koz, lot’s of very useful background there. I have one question on this: in a big club PA what is the low frequency that makes your chest almost vibrate/resonate?
As regards the high pass filter, many thanks for the tip on the cut off and roll off. Armed with this I went back and tried to apply that to one of my distorted files, but immediately realised why I originally abandoned that approach.
It should be apparent that I have little formal understanding of audio engineering, but I do have a little experience of digital signal processing from another field; it is rightly said that a little experience is a dangerous thing, but what concerns me applying the high pass filter is that it seems to raise some of the peaks in the waveform.
What I mean by this is that looking at say a 5 minute track in a compressed window with a flatish ceiling, following the application of the high pass filter this will become visibly jagged with the new peaks exceeding the old ones.
As I understand it a band pass filter should only subtract and never add to a signal, therefore I can not explain this behaviour and am distrustful of the result (in contrast the EQ works as I would expect).
Is there something I am missing such as the right filter quality q setting (BTW how should that be determined???) or is there something else going on I am missing?
Again many thanks to everybody for the wonderful help.
The low/high pass filters work in a similar way to (hardware) analogue filters.
An attempt at an over-simplified explanation:
As the corner frequency is approached there is a gradual phase shift. The result of this is that the peak of a particular frequency components my be in a slightly different place from where it was originally. This is the normal behaviour for this type of filter. Depending on where the peaks of specific frequency components occur, the combined amplitude of all frequencies at a particular point in time may be higher or lower than the original level.
The Equalizer effect uses a different type of filter (an FFT filter) that does not shift the phase, so the peak level will remain as expected.
Within the audio range, both types of filters produce sonically good results, (though FFT filters are unable to produce steep filter slopes at very low frequencies without an impractically large FFT size).
Thanks very much Steve, that is very helpful indeed.
I get how an analogue filter (emulation) wouldn’t produce the same results as a purely digital manipulation like Fast Fourier, and now feel comfortable in accepting the outcomes on your say so.
From my old clubbing days I always though Turbosound had some of the best sounding systems around, and a quick scan of their site seems to show their subs go down to around 25Hz.
So I’ll try the high pass filter again with that as the cut-off, and would like to use Koz’s suggestion of 24db as the roll off, to break the back of that possible hump on the spectrum I posted above around 12-15Hz.
They also make some great out-door rigs.
The low frequency range varies depending on the model, but for big subs it’s usually somewhere in the region of 25-40 Hz. Anything that goes below 20 Hz is an engineering research tool rather than something for entertainment (though vibrational stress testing can be quite entertaining )
No. the 24 dB roll-off does not use the Q setting.
The only rolloff setting that uses the Q setting is “12 dB per octave”.
All other rolloff settings ignore the Q value.
(This is why I think the setting is confusing and should be removed.)
When rolloff is set to 12 dB per octave, the default (Q = 0.7071) is the optimum setting and produces the expected response for a 12 dB/oct filter (-3 dB at the cutoff frequency then 12 dB/oct).
If the Q is set higher than the default, there is a peak in the filter response just before the set frequency.
If Q is less than the default then the filter “knee” is more gradual so the cutoff begins before the set frequency and is less than 12 dB/octave.
Steve, I am really, really grateful for the time you are taking on this, it is much appreciated.
OK, I’ll take your word for this. But when I compare the results of the 12db roll off with q=0.7071 versus a 24db roll off (and the same q you say does nothing) there appears to be a materially greater resulting magnitude in the new waveform peaks using the latter settings.
Based on your examples I would have expected them to be roughly the same if the Q implied in the latter calculation were optimised, and as such it could appear this is analogous to the situation you show where Q is greater than the optimum ‘default’.
However, I’ll just put this down to the vagaries of the analogue style filter, if you’re sure there is no error in the internal calculations.
It perhaps does look a little unexpected, but actually produce a very close match with the results that would be predicted for this type of filter. The mathematics are rather complicated (over my head), but if you want to look it up, these filters are “biquad filters” (biquadratic) http://en.wikipedia.org/wiki/Digital_biquad_filter
Here’s an illustration of the phase shift that occurs.
This is a 1kHz sine wave. The first track is unfiltered and the other tracks are each filtered with a high-pass filter. The rolloff is set to:
Track 2: 6 dB/oct
Track 3: 12 dB/oct
Track 4: 24 dB/oct
Track 5: 36 dB/oct
You can see how, as the steepness (number of dB’s per octave) increases, the phase waveform is gradually shifted backward (to the left)
The actual amount of phase shift depends on the frequency of the waveform being processed and the filter settings being applied. Because some frequencies are shifted more than others the frequency components in a sound will combine differently after filtering to produce different (a bit higher or lower) peak levels at any specific point in time. Human hearing is largely insensitive to phase shift, but is highly sensitive to frequency content, and the overall effect of these filters is to produce a precise and predictable change in the frequency content (as illustrated by the spectrum plots in my previous post).
I don’t understand the mathematics well enough to say if the phase shift is exactly as the theory says that it should be, but the frequency response is certainly a very good match with an idealised filter.
While the ear doesn’t respond well to phase problems, the effect isn’t zero. Sound system with low phase shift are frequently described as more “open” and “clear” and less “restricted.”
This is also the reason square wave testing keeps going into and out of favor. It goes into favor when people find it can accurately describe their equipment shortcomings. It goes out of favor when the manufacturers find it can accurately describe their equipment shortcomings.
And just to be clear, Fourier described the number, frequency and phase of different pure sine waves need to make up any other waveform including, notably, square waves. In the best case of “everything is hooked to everything else,” mess with the response anywhere and the effects ripple all over. Particularly since FFT is Best Guess. It’s not perfect.
That’s a really vague statement most likely to be found in audio equipment adverts.
Phase shift within the crossover of a speaker system can cause comb filtering due to the phase difference between the speaker drivers (tweeter and woofer) which will interfere with each other in the crossover frequency band. One of the things that make PA speakers by Turbosound so good is that they have gone to great lengths to ensure that all the speaker drivers are in phase with each other.
Phase shift is often deliberately introduced in high quality live sound reinforcement systems to avoid acoustic feedback.
Steve, I think I’m a bit like you in that when I see a complex looking equation my eyes glaze over and the mind goes numb.
Considering the entire discussion I am therefore more inclined to revert to the 12db roll off - firstly because the parameters seem more clear cut and secondly to minimise the new peak amplitude - but instead up the cut-off to maybe 30Hz or even a little more.
I keep turning Koz’s comments about the lower frequencies I am capturing not coming from the vinyl. In understand that position, but I can see many waveforms recorded with the same setup up where the spectrum falls off the chart at around 20db as expected, which seems to suggest it is not the overall equipment setup.
Secondly the waveforms with all the action at the low frequencies look similar regardless of whether I am sampling for the spectrum at the start or the end of tracks suggesting to me some external environmental factor is not the cause.
However, just by guesstimation it seems to me that these issues are largely cropping up on 90s dance type records. Is it not possible that associated - shall we say - informal methods of production and mastering eg in the bedroom are responsible for this?
In any case it might be wise to get rid of more low frequency stuff in the frankly unlikely event I ever again get to spin some of these tunes on a monster Turbosound system
Is it not possible that associated - shall we say - informal methods of production and mastering eg in the bedroom are responsible for this?
Before the RIAA forced the adoption of one vinyl customization curve, everybody had their own and each one was far better than the others. I wouldn’t be shocked to find dance records tailored to the Stanton 680 phono cartridge (for example).
It sounded OK, was hard to break, and you could back cue with it. That back-cue talent caused it to have “restrained” low frequency response.
And everybody knew that. The SL1200 was well damped and had all those ropes and pulleys (or modern equivalent) that would keep the needle in the groove in the face of a small, intimate thermonuclear event, so it was open season on bass music.
Vinyl cutters have “look ahead” to make sure that bass-heavy grooves don’t damage each other. They widen the cutter groove pitch before the music gets there. The joke was Disco/Trance/Electro disks had one song on each side and the grooves were wide enough to lose your car keys in there.