For those who might be interested, I have prepared time-domain signal capture of horn and violin, from CD output. One can see limitations of the CD format. The sample rate of captures was 100 samples in one time div, i.e. from 500kS/s to 10MS/s.
Please read information about time/div at left bottom corner of the images.
http://web.telecom.cz/macura/minstruments.htm
Please read information about time/div at left bottom corner of the images.
http://web.telecom.cz/macura/minstruments.htm
I can argue either side of the fence on this one. Well recorded CDs can sound great, probably as good or better than well recorded vinyl. They cannot, however, sound as good as a digital recording at a higher bit rate and depth. It wasn't clear what your original signal source was, but you can probably prove this with a decent condenser mic and interface on your PC. I haven't done that exact comparison, but the sound quality of recorded live instruments or vocals, and no processing, is beyond anything you ever hear on commercial CDs. I can also hear a difference when I transfer LPs to CD at 44.1 vs. 96 kHz. Small, but it's there.
Vinyl has it's own set of problems, but IMO these can be turned to our advantage as tuning tools. You can fool with cartridge type, stylus shape, VTA/SRA, energy absorption under the record (different types of mats), tonearm damping, and probably a dozen other variables. If everything is properly balanced, vinyl can sound darned good. The noise issue is minor on a good pressing, approaching inaudibility. Alas, the used records I buy are another story. A CD player, OTOH, only changes its character by a limited amount, regardless of the modification applied. Given the low source quality these days, I'm amazed that anyone spends more than a dollar ninety eight on a CD player.
Unfortunately, we can't record a high quality LP as we can do with a CD. If we could, and could do the kinds of signal comparisons we do for digital recordings, I think everyone would be pretty horrified at how badly LPs distort, and how adjacent groove modulation affects things, and how energy transmission through the vinyl causes ringing and artifacts. IMO, properly balanced, these things are what gives vinyl a certain sense of liveness and reality. IOW, the flaws improve the sound, as odd as that seems.
My take is that a better digital system would be great, but big changes would have to happen in the studio and subsequent processing, for it to make any useful difference. Right now, the source material is the limiting factor, not the bit rate or depth. I think they actually did a better job and targeted a more discriminating customer in the heyday of LPs.
Vinyl has it's own set of problems, but IMO these can be turned to our advantage as tuning tools. You can fool with cartridge type, stylus shape, VTA/SRA, energy absorption under the record (different types of mats), tonearm damping, and probably a dozen other variables. If everything is properly balanced, vinyl can sound darned good. The noise issue is minor on a good pressing, approaching inaudibility. Alas, the used records I buy are another story. A CD player, OTOH, only changes its character by a limited amount, regardless of the modification applied. Given the low source quality these days, I'm amazed that anyone spends more than a dollar ninety eight on a CD player.
Unfortunately, we can't record a high quality LP as we can do with a CD. If we could, and could do the kinds of signal comparisons we do for digital recordings, I think everyone would be pretty horrified at how badly LPs distort, and how adjacent groove modulation affects things, and how energy transmission through the vinyl causes ringing and artifacts. IMO, properly balanced, these things are what gives vinyl a certain sense of liveness and reality. IOW, the flaws improve the sound, as odd as that seems.
My take is that a better digital system would be great, but big changes would have to happen in the studio and subsequent processing, for it to make any useful difference. Right now, the source material is the limiting factor, not the bit rate or depth. I think they actually did a better job and targeted a more discriminating customer in the heyday of LPs.
Conrad Hoffman said:
I think that the flaws (in a live recording) allow you to place yourself in the hall.
This is evident when they play the recorded live performances of the Met Opera.
Back in the early 1990's I visited Aureal Semiconductor -- they never made any money except when sound cards first came into vogue -- but the guys who ran the company were into electronic music and had gone the synthesis route in the time domain. They had analyzed brass and string instruments, the effects of a larger bell on a trumpet for instance, and had from this created virtual instruments which would have been physically difficult to implement in the real world.
Regrettably, the got crushed by Creative Labs -- Creative sued everyone including those who weren't infringing on their IP.
For a comparison, a horn on vinyl:
http://web.telecom.cz/macura/vinyl_horn1.gif
http://web.telecom.cz/macura/vinyl_horn2.gif
Measured at 2.5MS/s sampling rate. Have a look at the rate of signal change, this is impossible to record on a CD.
(Miles Davis, Columbia)
http://web.telecom.cz/macura/vinyl_horn1.gif
http://web.telecom.cz/macura/vinyl_horn2.gif
Measured at 2.5MS/s sampling rate. Have a look at the rate of signal change, this is impossible to record on a CD.
(Miles Davis, Columbia)
I see some stuff up at 16kHz. That doesn't seem so difficult. The two questions I have are, can you hear it, and what does it look like after being sampled at 44.1kHz, *at the analog output* of a decent CD player. I've been fooled looking at raw data that's scary, then seeing and hearing a tolerable signal on playback. I'd also add that I hear differences between the software drivers on different PC media players, and that's completely different compared to my CD player, so I'm not sure the HF behavior up where I can't hear it, is significant.
It is much more than 16kHz. You can see edges shorter than 20us, followed by quite sharp change of slope. This indicate to frequency spectrum far above 20kHz. Reconstruction filter of the 16/44.1 would unify this by its impulse response. The resulting signal at CD output is convolution of samples with impulse response. Compare step response of the CD (same page) with horn2 sampled signal. Decent or not, 16/44.1 format always has sharp 22kHz digital brickwall in the chain.
Well, it is significant, as you can hear when you change linear PCM Fs to 96kS/s or 192kS/s.
Well, it is significant, as you can hear when you change linear PCM Fs to 96kS/s or 192kS/s.
Few years ago I made a FFT analysis of that Miles Davis LP track. The result is here:
http://web.telecom.cz/macura/miles.jpg
As you can see, the frequencies between 20 and 30kHz are well seen, and there is a spectral content up to 40kHz. CD would sharply cut everything above 22kHz, and brings its typical technology, unnatural sound.
http://web.telecom.cz/macura/miles.jpg
As you can see, the frequencies between 20 and 30kHz are well seen, and there is a spectral content up to 40kHz. CD would sharply cut everything above 22kHz, and brings its typical technology, unnatural sound.
Well, I'm just not seeing what you're seeing in those graphs. The FFT maybe, but it could as easily be distortion products that far down and that high up. LPs are anything but clean in a pure technical sense. If you really want to get a baseline for what's needed, play some notes or find someone to do so, and record that. Do same analysis. Look at the filtered reconstructed signal, not the data files. Getting from what you see in a waveform, back to why something sounded a certain way, is fraught with problems; I keep my confidence level on such matters purposely very low.
Conrad Hoffman said:
This has been already done many times, e.g. here:
http://www.cco.caltech.edu/~boyk/spectra/spectra.htm
The main problem of 16/44.1 is the low cuttoff frequency of the brickwall filter, which is necessary for proper digital signal processing to cut aliases and mirrors, and unfortunately affects natural musical signal.
I did look at the reconstructed filtered signal, I hope you got it. I sampled the CD output analog signal at about 50 - 100 x higher rate than is the original sampling frequency. What I saw is that the output signal bears a signature of brickwall filters used (Fs/2), and this indicates to unsufficient sampling frequency, which is the main problem of the CD format.
Can you transform signals from time domain to frequency domain? If not, then do not say that you do not see in the graphs what I do see.
Now that (transients) I agree with, but keep in mind that time domain measurement have provided little hi-fi enlightenment over the last 50 odd years, and everybody went to spectral analysis hoping to do better. It hasn't really happened, and consensus on what's audible is still sadly lacking. IMO, you need both, but once you've presented data, some correlation with audibility is needed. The bottom line is that so far the differences between well recorded 16/44.1 and higher bit rates/depth, have proven minimal. Not zero, but not the great improvement everybody hoped for. Everybody's looking for the "X factor", but there's a huge body of evidence that says it doesn't exist, and audio quality can be summed up with very conventional thinking, and within the normally accepted hearing band of 5-20 Hz, up to maybe 20 kHz, depending on the individual. That applies if you're under 30 or so. If you're older and can hear 16 kHz, you're doing well, and as someone in another post mentioned, response typically drops off like a stone off a cliff. Dozens of dB in a few hundred Hz. If I were a bat, I'd starve.
Sadly I must say, if you hear a cymbal crash and it does not contain all the harmonics with the right ampliftude, then it does not sound like a cymbal but like a shoe box. with 44.1 kHz sample rate you hardly have two samples of 20 kHz and it may not have been taken at the right instance so you may only have two samples of 10kHz if you are lucky.
This has been a problem all along with CD, people who play real accoustic instruments have a problem resolving the charater of the instrument. This has also been proven over and over. Although 24 bit has a higher amplitude resolution and dynamic range, the sample rate fails dismally.
This has been a problem all along with CD, people who play real accoustic instruments have a problem resolving the charater of the instrument. This has also been proven over and over. Although 24 bit has a higher amplitude resolution and dynamic range, the sample rate fails dismally.
Nico,
this is just the case of us who are familiar with live acoustic, unamplifyied instruments and concerts. The people who know only reproduced electronically amplifyied and recorded music do not understand the issue and often come with trivial engineering attitude. The further case is just commercial - MP3 and 16/44.1 is enough then.
Regards,
Pavel
this is just the case of us who are familiar with live acoustic, unamplifyied instruments and concerts. The people who know only reproduced electronically amplifyied and recorded music do not understand the issue and often come with trivial engineering attitude. The further case is just commercial - MP3 and 16/44.1 is enough then.
Regards,
Pavel
Nico Ras said:
The frequency resolution of spectrum analysis is sampling frequency divided by number of samples in the record. So, for 96kHz sampling and 64K samples you get resolution like 1.46Hz. But you need those 64K samples, and there can be few transients between them. 64K samples of 96kHz sample rate takes 0.68s of time, this is a long time, and FFT is averaged in this time interval. You can have several fast transients during this interval, that are "hidden" in FFT, but can be easily captured in time domain.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.