I've been reading these forums throughout the summer and you all, along with several online guides, have helped me tremendously with what I would like to do with my music. I take great pride in my CD collection, and after the purchase of a big hard drive, I'm in the process of ripping hopefully all or as much as I can fit losslessly onto the drive. I'd really like to be able to participate in some of the discussions here, but frankly, you all seem incredibly knowledgeable and are light-years ahead of me in the subject. I can follow along with the guides (configuring EAC, foobar, etc.) easily enough but there are still many terms and subjects I have no clue about. I looked at the 'Technical' section in the wiki but I have no idea what context many of those terms fall under. This topic is broad because I have no idea where to begin. Other threads advise reading about DSP but I was wondering in general where to start, what to read, etc.? All of that. Is DSP truly the place to start? Where did you all get your info, and more importantly, where'd you begin? It's mighty intimidating to an absolute newcomer such as myself, but I'd really like to learn.

Thanks for your time.

I suggest asking yourself, "What's missing in my music?" and taking it from there.

Is this in the wrong forum or something? I have no idea what you mean.

Do you feel something is lacking in the way you enjoy your music? If so, what's getting in your way of enjoying it? Starting is difficult without goals.

Put short and sweet, I want to learn about digital formats. Codecs, how they work, where these formats are headed, artifacts (I still have no clue what they are), etc. The works. I know this is terribly vague, but I really don't know much of anything on the subject and I have no idea where to begin.

Here's a start:

ff123's Artifact Training Page

Thank you!

1. Digital audio is most commonly PCM.
2. You compress PCM either losslessly or lossily.
3. Lossy compression is usually psychoacoustic.
4. Psychoacoustic compression works because your ears are not microphones. Even if two sounds are not the same, they might sound the same to you!

Can someone write something brief and helpful about time domain and frequency domain, please?

1. Read more at http://en.wikipedia.org/wiki/Pulse-code_modulation
3. Lossy compression uses psychoacoustics to determine how and which information in PCM can be reduced to gain as much compression as possible or desired.
4. Read more at http://en.wikipedia.org/wiki/Psychoacoustic_model

I hope this will help a bit to clarify things.

Once when you learn more about artifacts, maybe you can join and contribute to a listening test http://www.hydrogenaudio.org/forums/index....c=56397&hl= You can also discover artifacts yourself while you are doing the listening test

Guess you want to dig a bit deeper and not just follow blindly to advices. Roughly speaking that's what most members do here. But don't think you have to dig extremely deep: codec developers (or members with a comparative deep knowledge) are rare here (though highly appreciated).

As for a start I would forget about DSP details right away. In case you'll find yourself burning for such details you can do that later.

For an understanding of codecs basics my advice is to learn about

• basic digital representation of music (get to understand the essence - not every detail - of PCM, WAV, sample frequency, sample accuracy). This should be easily achieved.
• Technical mp3 basics as these basics apply to many codecs as well (get to understand that an mp3 data stream is build up of frames each corresponding to a certain number of wave samples, that frame size corresponds to the frame's bitrate, that each frame can have its specific bitrate out of a pregiven list of frame bitrates, that constant bitrate means that each frame bitrate is the same while with variable bitrate the bitrate of each frame can be different, that the encoded data of the wave samples associated with a specific frame can be stored also in neighboring frames within certain restrictions [the concept of bit reservoir] which effectively means audio data bitrate is varying even in the case of contant frame bitrates, learn about stereo representation (apart from left and right: mid (left+right) and side (left-right) as well as using any pair out of these for stereo representation [joint stereo], that encoding the data for the wave samples envolves splitting the music into different frequency ranges (learn about critical bands), analysing these for masked signals that can be ignored, and get a rough understanding of how these critical bands are encoded. Learn a bit about psychoacoustics used in skipping the nonaudible information).
  This will take a while. The internet is full of information and you can inform yourself to any level you like (though it will not always be easy). Keep in mind: you don't have to know this all, especially not in detail, so stick with the topics most interesting to you.

There's a lot more depending on your interests, but that's what I would start with.

Time domain: That's our usual understanding: wave signal as a function of time.
Frequency domain: Any wave signal can be alternatively represented by its frequency spectrum. Think of a specific note played by a certain instrument. This consists of a fundamental note and harmonics, and their frequencies, amplitudes and phase relations make up for the frequency spectrum of this specific note played. As for a single played note it's for any wave signal.
So for a certain small time window the frequency spectrum can be approximated and encoded. Due to the time window and an in practice restricted approximation to the frequency spectrum there is a limited temporal resolution with the frequency domain representation. Luckily there's temporal masking in psychoacoustics, but if the time window is not small enough (and/or the representation of the frequency spectrum too poor) the limited temporal resolution is audible as pre-echo.

IMO, that's a bit misleading.
Frequency-domain and time-domain representations of a signal are absolutely equivalent and can be converted perfectly, losslessly -- using any "time window", narrow or wide window it doesn't matter.

So it is the "approximation to the frequency spectrum" (and on top of that also overlapping of the windows) what limits the temporal resolution.


So, it is not "and/or" it is strictly "and".


/Edit: oh, and by the way, DSP was not my major, so anyone please feel free to correct me.

@kdo:

You're right: The primary problem of temporal resolution is that of a bad approximation to the frequency spectrum.
You're wrong: Concerning temporal resolution a bad approximation has a lot to do with the size of the underlying window (given a certain amount of more or less fixed computational resources).

Could you please explain what do you mean? (Maybe we are talking about different things? )

What I mean is very simple: Take a signal, slice it into several chunks in time-domain. Let's call each chunk a "window". Apply one of the "classical" transforms (e.g. plain Fourier) --> we get frequency coefficients, for every chunk.
Now, let's keep all these frequency coefficients (no approximation, in other words). Then the inverse transform can be applied on these coefficients --> we get our original signal back, exact. Lossless.

Of course, that's not how lossy encoders work --> there is always some approximation --> and only when there is approximation the size of the window matters.

That's what I'm talking about. Cause we always do approximate.

Maybe the misunderstanding arises from the fact that in the first place I answered kjoonlee's question in a general (more theoretical) sense.
With my last paragraph I addressed approximation considering the practical aspects especially with respect to restricted temporal resolution. Thought that was clear.

But I don't like to continue this cause I think we actually do talk about the same thing. Presentation of a certain subject will always depend on the one who presents it.