Sound and the Web

When you're compressing sound, the first step is to leave most of it out. While that may seem surprising, telephone engineers have known it for years. You can slice out most of the highest and lowest parts of the audio spectrum, and average listeners probably won't notice any quality degradation. (This is mainly true for spoken sound, not for music, which is why you find very few bands recording albums over the telephone.)

Another nice feature of sound, for the purposes of compression, is the large degree of repetition found in most speech and music. When you're sampling music 44,100 times a second (for CD quality), much of it can be encoded as "ditto." Same for speech: Vowel sounds vary little from beginning to en d, and there are not that many consonantal sounds to encode (which makes you wonder why all artificial speech programs sound like inebriated frogs).

Not surprisingly, there are many more products for sending speech (i.e., telephony) over the Web than there are for music. Sound compression is quite lossy, and most people can tell the difference (that's one reason we listen to FM radio rather than AM and CDs rather than records).

Music over the Web requires higher bandwidth than even a 33.6-Kbps modem connection. Once you have tossed out the frequencies you don't want, there are many choices for compressing the result, including Lempel-Ziv-Welch (LZW). Because compressed sound travels as packets over the Internet, you must plan for packets going astray. One strategy, implemented by VocalTec's Internet Wave, uses the Transmission Control Protocol (TCP) with a buffer that stores a few seconds of sound, plus a predictive cache. The buffer provides a few seconds' wait for any dawdli ng packets to show up, and if the packets never show up, the prediction algorithm guesses at their contents.

Another strategy, one that's implemented by Progressive Networks' RealAudio, employs the User Datagram Protocol (UDP), with the sound minced and distributed among successive packets. If one of these packets does not show up, the effect is a few-milliseconds gap every quarter of a second for about 3 seconds.

Want to pause, fast forward, or rewind? Unfortunately, HTTP is not bidirectional. Again, there are ways around this: VocalTec, for example, uses Common Gateway Interface (CGI) scripts with HTTP; Progressive Networks uses other non-HTTP servers.