|
|
 |
DATA COMPRESSION CONCEPTS
To deliver content to a user involves a transfer of data, whether that data be text, audio, video or graphics. There are two sides to this coin. Larger bandwidth delivery methods (ie. Broadband or faster modems) and data compression.
In dealing with audio transfers we need only look at the size of full bandwidth CD audio to quickly realize that data compression is necessary. In order to stream mono CD quality audio in real time - that is to push the file over the 'net to the user's computer so fast that it plays continuously without having to download first - requires a bandwidth of over 700 kilobits per second. This format is what's known as linear or PCM audio and is not generally considered to be data compressed. The audio signal is "sampled" 44,100 times per second and the resultant value is documented using 16 bit words to store the samples, this is commonly called 16/44 audio. The bitrate of this (that is the number of kilobits of data per second - similar to the baud rate of a modem) is 16 X 44.1 or 705.6kbps. Since the majority of web users are still using 56K modems, which actually yield only about 40 kilobits per second of throughput, you need to compress the file size to stream audio.
Many CD-ROM games use Wave or AIF files (both linear formats) at 8 bit/11.025khz (8/11) which is roughly 88kbps. In a CD-ROM the concern is disc storage space, not bandwidth of course. This is called "lossy" compression. The loss is noticeable in the sound quality. By reducing the sample rate to 11k, the high frequencies are greatly reduced. By reducing the word size to 8 bits, the noise becomes greater, resulting in a crackling and/or hissy sound.
Fortunately there are now many ways of reducing file/bandwidth size while retaining higher quality sound. This is commonly called "lossless" compression (although there really is a loss of quality, but much less noticeable then simply reducing the sample rate and bitrate.
For music files MP3 is the norm these days, typically using 64kbps per channel, or 128kbps for Stereo audio. MP3 and all other types of audio (and video) data compression use intelligent methods of eliminating unnecessary information. The software used for this is called a codec - for encode/decode, similar to a modem which comes from modulate/demodulate. The sound files are encoded by the developer and decoded in real-time by the helper application. Audio can be encoded in real time as well for live webcasts and other applications such as Internet Telephony.
For speech files, bitrates as low as 6.5kbps yield intelligible and surprising low noise audio files. RealAudio and QuickTime have some of the best speech quality codecs in use.
The audio in the ItSpeaks Initiative is encoded using 24kbps MP3 as audio-only Flash files (see Adding Audio-Only Flash Files to HTML Based Sites.) To us this is the best compromise between sound quality and size. We chose Flash because of the widespread installation of the Flash Player in most people's browser.
This audio will play over a 56K modem, with only the occasional stop and start, and noticeable, but acceptable quality loss. It also allows for music and sound effects to be mixed with the voice, these actually seem to help the quality of the narration.
As the Initiative grows we plan on including samples of various formats.
|
|
 |
|
