Enter Bitrates, Stage Left
While MP3 users cannot control the degree of lossiness specifically, as they might do with a JPEG image, they can control the number of bits per second to be devoted to data storage, which has a similar net result.
In the process of coding, the "irrelevant" portions of the signal are mapped against two factors: a mathematical model of human psychoacoustics (i.e., the masking requirements), and the bitrate, which is established at the time of encoding (see Chapter 5). The bitrate simply refers to the number of bits per second that should be devoted to storing the final product-the higher the bitrate, the greater the audio resolution of the final product, as shown in Figure 2-3. An easy way to visualize the effect of bitrate on audio quality is to think of an old, turn-of-the-century film. Old movies appear herky-jerky to us because fewer frames per second are being displayed, which means less data is distributed over a given time frame.
Figure 2-3: More bits per second means more audio resolution, pure and simple
For example, the current de facto standard is to encode
MP3 at 128 kbps, or 128,000 bits per second. The codec takes the bitrate into consideration as it writes each frame to the bitstream. If the bitrate is low, the irrelevancy and redundancy criteria will be measured harshly, and more subtlety will be stripped out, resulting in a lower-quality product. If the bitrate is high, the codec will be applied with leniency, and the end result will sound better. Of course, the
file size of the end product corresponds directly with the bitrate: If you want small files, you have to settle for less quality. If you don't mind larger files, you can go for higher bitrates.
Bitrates refer to the total rate for all encoded channels. In other words, a 128 kbps stereo MP3 is equivalent in size and quality to two separate 64 kbps mono files. However, a 128 kbps stereo file will enjoy better quality than two separate 64 kbps mono files, since in a stereo file, bits will be allocated according to the complexity of the channels. In a given time, one channel may utilize 60% of the bits while the other uses only 40%. The cumulative size in bits will, however, remain constant.
CBR vs. VBR
Most of the information you'll read in this book and elsewhere assumes that the bitstream is being encoded at a
constant bitrate (CBR). In other words, if you specify a 128 kbps encoding, then that's what you're going to get, start to finish. The drawback to CBR is that most music isn't structured with anything approaching a constant rate. Passages with many instruments or voices are succeeded by passages with few, simplicity follows complexity, and so on. The response to this situation has been the development of
variable bitrate (VBR) encoders and decoders, which vary the bitrate in accordance with the dynamics of the signal flowing through each frame. VBR technology was first implemented by Xing, which is now owned by Real Networks, but is now supported by dozens, if not hundreds, of third-party products.
Rather than specifying a bitrate before encoding begins, the user specifies a
threshold, or tolerance, when encoding with VBR. All notions of bits per second go right out the window, of course; instead, one selects VBR quality on a variable scale. Confusingly, this scale is represented differently in different encoders. While MusicMatch Jukebox gives you a scale of 1 to 100, the LAME command-line encoder lets you specify a quality of 0 to 9, where the scale represents a distortion ratio. Therefore, you can't just assume that higher numbers mean higher quality-see the documentation for your encoder before proceeding, or run the tests yourself. In any case, the scales are essentially arbitrary; think of them as though you were using a slider to control the overall quality versus file size ratio as you might with a JPEG editor.
While VBR files may achieve smaller file sizes than those encoded in CBR at a roughly equivalent fidelity, they present a number of drawbacks of their own. First, these files may not be playable in older-generation decoders, which had no notion of VBR concepts (although the
ISO standard specifies that a player must handle VBR files if it's to be considered ISO-compliant). Second, VBR files may present timing difficulties for decoders. You may expect your
MP3 player to display inaccurate timing readouts-or no timing information at all-when playing back VBR files. However, VBR techniques conveniently take some of the guess work out of trying to find an optimal bitrate for any given track-whereas you might have to encode a file several times with CBR to find the perfect balance, you can just set your encoder to use a relatively high quality level and let the computer figure out an optimal bitrate for each frame automatically.
In general, the header data in most
CBR files is same for each frame, while header data is necessarily different for each frame of a VBR file. However,
VBR files don't incur more processing power, as all MP3 players read the header data for each frame regardless of whether they're playing a CBR or VBR file.
Bitrates vs. samplerates
Bitrates aren't quite the final arbiter of quality. The resolution of audio signal in general is in large part determined by the number of source samples per second stored in a given format. While bitrates are a measure of the amount of data stored for every second of audio,
samplerates measure the frequency with which the signal is stored, and are measured in kiloHertz, or thousands of samples per second. The standard samplerate of CD audio is 44.1kHz, so this is the default samplerate used by most encoders, and found in most downloadable
MP3 files. Audio professionals often work with 48kHz audio (and, more recently, 96kHz). Digital audio storage of lectures and plain speech is sometimes recorded as low as 8kHz. Streamed MP3 audio is often sent out at half, or even a quarter of the CD rate in order to compensate for slow Internet connection speeds. If you need to minimize storage space, or are planning to run your own Internet radio station, and are willing to sacrifice some quality, you'll want to do some experimenting with various samplerates. More details can be found in Chapter 5.
Note that nothing is ever actually played or heard during the encoding process-you can encode MP3 on a computer with no sound card or speakers, if you need to for some reason. In fact, this is exactly how things are done in some professional organizations, particularly those dedicated to Internet broadcasting (see Chapter 8, Webcasting and Servers: Internet Distribution). In such instances, one computer may be used for auditioning and selecting files, a second used for the actual encoding process, and a third dedicated to serving the files to the Internet. Of course, the beefiest machine available will always be used as the encoding machine in such a scenario.
Next: Freedom of Implementation