Improving CD 44.1kHz/16bit audio

[Mr. A]

Hi everyone. I'm new to the forum and to Audacity. I found Audacity when I searched for a program that could simply chop and edit some music for a project I'm putting together. Needless to say I got a lot more than I expected. I have possibly an unusual question about some things I've been experimenting with on this software.

I was recording some music off one of my favorite music DVDs and my attention was drawn to the left where it lists the bit rate and sample frequency. I then decided just for grins to up the bit rate to 32 bit and the sample to 96kHz and recorded the DVD in real time again and exported it to a WAV for CD playback. I found that the higher bit rate and sample sounded a great deal better. I then wondered about my CD library. To make a long story short, I have been importing files ripped by Windows Media Player, up sampling them, then slowing the speed down to %53.990 to counter the effects of the up-sample and exporting it to a wav and being astounded in the increase in sound quality.

Now I'm am new to recording programs, but not new to sound. I am a HT/music enthusiast that stumbled on this program. It seems illogical to me to import a lossy cd track, up-sample and convert it back to a 16 bit WAV and it sound noticeably better. By better I mean the music sounds more transparent, more realistic, more accurate. It doesn't sound brighter or warmer like a high frequency boost or cut. It's not just one thing that improves, it just all sounds better. I've played this music back on my computer speakers, car, and music system and all yield significant improvement. My unusual question is: How is this possible? The file size is the same as the original and the wavform is 16 bit. I've had people on other forums say that I'm actually introducing errors in the original content. However if what I am doing is wrong, I don't wanna be right. If I am F'ing up the original and it comes out sounding better, I'll F up my whole music collection.

So why do I care? Why don't I just do this thing and be happy and not question it? Because I'm a tech geek and cannot be happy with just knowing that it works. I wanna know what it really going on to make some sense of all this. Sorry for the long post and I'll appreciate any responses because I am so lost.

[kozikowski]

I'm with the distortion people.

But that's not a bad thing. People buy Neumann large capsule condenser microphones not for their accuracy, they're just not that good, but they tend to fall madly in love with the human voice at every opportunity.

See: Tube Microphone Preamplifiers. Still used. Still sold. Still distorted.

Know anybody willing to give up their vacuum tube Fender guitar amplifier? Why?

I spent a lot of money for a Grado phono cartridge. I bet I can prove that its accuracy isn't that good, but it makes my vinyl sound wonderful.

Go with it.

Koz

[kozikowski]

You might be falling into the Great Nyquist Debate about Music CDs. 44100 can't really do accurate sound much above 15 KHz, FM Radio quality. They play enormous tricks to simulate frequency response above that. There are a number of people that can hear the tricks.

I wonder if the oversampling and resulting processing just gives you the same sound that the early very expensive CD players used to do. They had custom frequency contouring and tuning networks in there to avoid sampling errors and smooth out the presentation. I'm betting no $49 CD player has all that stuff in there.

I had at least one console CD player that featured oversampling.

Very early floor-standing videotape machines used to work by doubling the datarate and then decoding, so that does work.

Koz

[Mr. A]

I've been on the information super highway ever since your posts. Thank you for the responses. You mentioned the Nyquist term and I searched that out and learned a whole lot more than just about that. While still not making total sense of what is happening I now have a few theories.

[kozikowski]

When I went through school--with Edison--the Nyquist value was 2.6. You needed at least 2.6 digital sample points in any waveform to reconstruct the original analog signal from the digital data. Now, it's universally considered to be 2.0. The electronics didn't get any better, it was just recognized that all signals have noise and the noise (or dithering) is required for the lower Nyquist setting to twork. If you have a pure, noise free tone then you're back to 2.6 and slightly over 15 KHz response without distortion.

Koz

[Mr. A]

What do you mean by 2.6 sample points?

[kozikowski]

There is a pretty good explanation of digital sampling right here in the Audacity system.

http://audacityteam.org/manual-1.2/tuto ... ics_1.html

The third illustration down is a pure tone (like a flute). The dots represent the "samples" that the electronics used to digitally record the wave. Obviously, if somebody handed you just that pattern of dots (representing digital numbers), you would have absolutely no trouble connecting them one to the next to reconstruct the original flute tone wave.

Now do that same wave--same flute tone--with only three dots. If somebody handed you only three of the dots in the third illustration and didn't tell you what the music or the wave was ahead of time, could you put the flute tone wave back together?

Welcome to the Nyquist Limit.

Koz

[Mr. A]

Yeah I understand the sample points in the illustration, but when you say 2.6 sample points, is that per second, millisecond or what? 2.6 Per second is really bad resolution

[pdxrunner]

Required sampling rate is 2.6 times the highest frequency you want to resolve without distortion.
For example, 16.9 kHz * 2.6 = 44,100

[kozikowski]

<<<Required sampling rate is 2.6 times the highest frequency...>>>

What he said.

One of the tricks they pull is your complete inability to tell whether anything over 16.9 KHz is distorted or not. Given that you are a Super Being that can hear 16.9 KHz in the first place, in order to hear distortion, you need to be able to hear over that. Distortion damage goes up in frequency almost never down. Dogs, for example, can here distortion up that high.

So for the sake of your chihuahua, stay out of questionable Nyquist sampling. We note that some young women can do that, too.

Another trick they pull is the assumption that any mystery or scrambled sample points up that high must be musical tones because they can't be anything else. Then you work backward. What's the best musical tone that happens to fit the scattershot of sample points you have. This process goes into some pretty serious math that makes my head hurt.

Encoding CD shows, if you followed all this process, depends very much on not showing up to the party with sound that goes over 20 KHz. All these tricks go right in the toilet if you do that. Worse than creating distortion, the process creates new, clearly audible sounds that didn't used to be there.

Not a smooth road this Nyquist thing.

Koz