MP3 sounds fine to me, through my headphones at work, where the headphones and the soundcard are probably the main limiters of sound quality.

However, my local pub sometimes has a DJ who uses two PCs each running an MP3 player, instead of CDs or vinyl. One night, as the place filled up and became noiser, I realised I knew what song I was listening to, but I could not place the pitch; I wouldn't have been able to sing along because I wouldn't know what key to join in at.

Fortunately, I had an audiophile at hand to consult. He told me, there was practically no bottom end and very little top end -- my ears were missing vital information. Perhaps the pub's PA is lacking, but I suspect MP3 wasn't up to this demanding application; competing with the chatter in a noisy bar.

There is indeed already a node about this, fool.

Sound quality of an Mpeg 1 Layer 3 Audio File (hereafter 'MP3') is usually (in my experience) dictated by four things: Source, Encoder, Player and Output; quality is affected linearly in that order. I am not an Audiophile, and this is not written for an audiophile, just for people who like nice sounding tunage.

  • Source - If the audio you're grabbing is from a new CD, ripped digitally, you should not have any issues here. However, it's not totally foolproof. If the CD is badly scratched, your ripper will either simply record the scratch, or in some cases completely bork itself. Also, unless you have a God Box, using the PC while you rip is not a good idea since ripping a CD is very timing dependent -- just like burning a CD, if the flow of data it interrupted, the rip will end up with gaps and pops.

    If you're transferring from another source, though a sound input device, the quality if that input device comes in to play. Many sound chips are perfectly fine, but a number of older cards are crippled in the input department and only do things like 8-bit input or 22kHz sampling.

  • Encoder - Assuming the raw PCM Data made it into the PC without woe, the next step is to package it into the fabu MP3 format. Many encoders exist and, unless you're an audiophile, which encoder your use isn't an issue... within reason. Older Xing encoders should be avoided like the plague because they cut out at 16kHz versus the 20 or 22kHz that other encoders do; The newest Xing codecs do not have this problem because they are basically a copy of the Fraunhaufer codec.

    The Real Jukebox encoder has issues, too, so I'd recommend you avoid it.

    For most purposes a basic encoder like LAME should be fine, unless you're anal, in which case you'll want to look into something like the Radium codec or Encoder-X.

    Bitrate has always been a huge issue with most people, and so I'm not going to get into it in any great detail. Suffice it to say that, if at all possible, don't use 128kbps encoding. Very few encoders can handle complex wave forms or deep bass properly at 128, and so you're get mangled and clipped sound. Do yourself a favor and go to 160kHz, or even 192 is you can justify the extra disk space. Don't use VBR.. just don't.

  • Player - Once the file is encoded you, obviously, need a player to play it. Thankfully decoding an MP3 is much easier than encoding it, and so most MP3 Players do a decent (if not good) job of it. I'm partial to Winamp 2.71's Nitrane Decoder, with the stock XMMS decoder coming in at a close second. Players that suck include the old Micros~1 Media Player and Quicktime 4.0 (the decoder has been improved in 4.1 and 5.0).

  • Output - The last hurdle in getting great MP3 wave-forms is where the sound comes back out as analog sound waves. Sound cards are the usual cause of quality loss here, with speakers as a close second reason.

    While most sound chips are up to the task, many old sound cards only do 8-bit output and have horrid signal-to-noise ratios -- the original Sound Blaster Pro is a prime example. You might also be getting a 50/60-cycle hum in your speakers, usually due to improperly shielded power wires near your speaker lines. This can usually be remedied by using a Ground Loop Isolator. They're available from Radio Shock (in the US) and are specifically designed for this purpose.

    Another side not about output quality is the processing power of the device doing the decoding: Decoding a 44kHz 192kbps Data stream takes a lot of power, and if you don't have it you could be forced to drop your sampling rate to something like 22kHz. This is a bad thing, and if your can upgrade your computation capacity, do.

My (limited) understanding of MP3 compression is that it works on psychoacoustic principles. That is to say, when a track is ripped, it's spectral analysis information is compared to a psychoacoustic model stored in the encoder (what the encoder thinks you will hear). Any frequencies the encoder decides will be inaudible, are removed. This makes for great compression, but what about sound quality?

Some of the sounds an MP3 encoder removes probably don't make much difference to the average listener. Much of the time, though, the overall 'feel' or 'sound' of a song relies on VERY subtle factors (vinyl freaks know this, while CD enthusiasts seem to have forgotten it). MP3 compression removes a lot of these sounds. For example, if two notes sound at the same time, and one is much louder than the other, the softer sound is taken out. Also, as has been mentioned many times, the bottom end and top end are lacking.

As a result, casual listening (i.e. on headphones at work) is rarely affected. At high volume or on a good sound system, however, much of the warmth and expressiveness of a song is lost.

It's important to keep in mind that MP3 (or, correctly speaking, mpeg-1 level 3) is a lossy compression algorithm that relies on the way your brain and ears work. A good encoder (like lame) can do a wonderful job and generate a decent mp3 file, whereas a poor one (like bladeenc or any of those xxx-jukebox that use a xing encoder) will cut out high frequencies and introduce artifacts, annoying the hell out of you.

The most common problems with mp3 files:

Some guidelines I use when I create my MP3s:

  • Use lame as the encoder.
  • Use cdparanoia as the ripper. This is for linux. I've heard there are some very good ones for windows too.
  • 128Kpbs: only when I want to send a preview of a song to someone.
  • 160Kbps: a little better on the high-frequencies.
  • 192Kbps: A good trade-off, if you have a little bit more space to waste.
  • 256Kbps: Almost archive quality. Big files, but also good quality
  • Variable bitrate: The best deal for me. I set the quality to 2 (1 is the best, 7 is the worst), minimum allowed bitrate to 64Kbps and maximum to 256Kbps. With these settings, the encoder will vary the bitrate dynamically between these two values (it's actually fun to watch the bitrate changing while the music plays). The nice thing about VBR is that most songs have periods of silence that can be safely encoded at lower bitrates, saving space. Allowing the encoder to go as high as 256Kbps guarantees quality in some parts of the song. On the average, songs will stay around 180Kbps, but bits will be allocated where they're actually needed.
  • Use CRC on your MP3 files (that allows you to detect file corruption).
  • Use tags. They help a lot when you want to classify you collection.
Some caveats:
  • Some players don't like CRC information (I believe there's an AIWA car player that won't play MP3s with CRCs)
  • Some players don't like VBR.

That's not the trend though. I don't believe you should trade VBR and CRC error correction for compatibility with some obscure players.

Let us establish some solid facts as a basis for this matter. The SPL limit for human hearing is approximately a little less than 10 dB of volume in the most sensitive frequency ranges(the same ones used for speech). The frequency limit of human hearing is approximately 20 to 20 000 Hz. If you(yes, you) are using a regular $150 CD player, you probably have nothing better than a 20-bit DA converter connected to regular $300 speakers and an average 0,5% THD stereo amplifier built with ICs instead of discrete components. Its signal-to-noise ratio might be 70 dB or so. Of course a lot of people out there are audiophiles extraordinaire, but since they are already pissed about every part of the abovementioned stereo, they have bigger fish to fry than mp3.

Some people state that "when mp3's are encoded, specific frequencies are removed and discarded, frequencies that are said to be unheard by human ears." This is not true. If you look up a textbook on human hearing, you will notice that the sensitivity of your ears is not uniform throughout the frequency range 20Hz-20Khz. Instead, you have a lot better hearing in the middle of that spectrum. Mp3 uses this(and a lot of other physical facts, not just psychoacoustics) in order to compress music as carefully as possible. Few people complain about the bass in mp3s. Some people complain about the treble in mp3s. A lot of people complain about the midrange in mp3, especially about the elusive "warmth" of the sound. Snake oil aside, it is a scientific fact that this is the frequency range where human hearing is the most sensitive. It is also where the mp3 algorithm does the least of its compression. In fact, the test song used during the development of the mp3 algorithm was Tom's Diner by Susanne Vega.
why? Because Karlheinz Brandenburg, lead of the mp3 development project at the Fraunhofer Institute for Integrated Circuits, used it during testing specifically out of a need for "warm" vocals.

Some people refer to failures of mp3 compared to CDs. The mp3 algorithm has the unfortunate disadvantage of being based upon the abilities of the compact disk. Tomlinson Holman, who developed the THX standard and coined the term "5.1 channel", has suggested an ultimate goal of 10.2 sound, with an array of seven speakers covering the front of the room, plus three speakers in the rear area and two subwoofers, one on each side at 180 degrees.
Does it still sound like mp3 is the issue you have to worry about? Isn't it possible that a 256 kbit, 96 kHz, 10.2 channel mp3 ripped from DVD-audio might be a whole lot better in every way than CD?

Courtney Love has publicly stated that she thinks mp3s sound like shit. She did this in a speech where she also spoke at length about how bad the availability of alternative music on Napster is. Isn't it possible that the fact that most of the music on Napster is encoded by everyone and their dog at 128 kbit or les, downsampled to 22 kHz or ripped from scratched records might have something to do with it? Can you honestly tell the difference between a 192 kbit mp3 ripped by a dedicated mp3 piracy group and distributed with an .SFV file to verify data integrity and ripped from a new, unplayed and clean original CD? Can you do it even on the aforementioned $600 stereo?

One note on MP3 sound quality:

As a sound designer, I have discovered that in performance spaces with a lot of flutter echo, tracks recorded from MP3s tend to sound better than tracks recorded directly from CDs. This may be because of the 'inaudible' sounds that are removed during compression, leaving less sound to bounce off the walls. There is one theater I've worked in that is particularly bad when it comes to echo (concrete walls and nothing soft to absorb sound); in most cases, the flatter, muffled sound of the mp3 track is much better in the space than the sound from the original source, as the sound bounces so much it is almost impossible to hear clearly.

Log in or registerto write something here or to contact authors.