Web Performance Tuning

Audio

Most audio formats encode point-by-point air pressure over time using 8-bit samples, giving 256 possible amplitudes. This technique is known Pulse Code Modulation (PCM). Some formats distribute audio linearly, while others take advantage of the fact that humans hear in a nonlinear way. That is, humans have a harder time distinguishing two loud sounds than two soft sounds, so encodings are assigned more densely at small amplitudes.

All of the following formats use PCM in one way or another:

• Sun’s .au

• Microsoft’s .wav

• Apple’s AIFF

• mu-law (U.S. telephony)

• A-law (European telephony)

You can also code sound in the frequency domain; that is, the code says something like “play this frequency at this amplitude for this amount of time.” MIDI works like this.

The number of samples per second determines the frequency range you can encode: if you have n samples per second, you can encode frequencies up to n ⁄₂ Hz. (This is the Nyquist theorem.) The sample size and number of samples per second also determine the size of an audio sample, and therefore its download time as well.

In the telephone system, voice is encoded as 8-bit samples at 8KHz, giving an audio bandwidth of 4KHz and reasonable quality. Eight bits at 8KHz comes out to 64 kilobits per second, which is the bandwidth used in telephony between switches for a single voice channel. This is a fundamental limit on the rate of information that a modem can send on a single call over the voice network.

Police and fire department radio systems ...