6. Basics of Digital Audio

Audio data has special and unique properties: e.g. while it is useful to occasionally drop a video frame from a stream, we simply cannot do the same with audio information or all sense will be lost from that dimension. Therefore, how to sensibly compress sound information is an important question. We begin with a discussion of just what makes up sound, and consider the digitization of sound information. We introduce the Nyquist Theorem as a fundamental property of sampling. Signal-to-Noise Ratio (SNR) is defined and adopted as a useful measure of audio (and in general, signal) quality, including the effect of quantization noise. Linear and nonlinear quantization, including companding for audio data, are discussed. Synthetic sounds are introduced, and we then go on to a thorough introduction to the use of MIDI as an enabling technology to capture, store, and play back musical notes. We look at some details of audio quantization, and give introductory information on how digital audio is dealt with for storage and transmission. This entails a first discussion of how subtraction of signals from predicted values yields numbers that are close to zero, and hence easier to deal with Pulse Code Modulation (PCM) is introduced, followed by differential coding of audio and lossless predictive coding. Finally, Differential Pulse Code Modulation (DPCM) and Adaptive DPCM are introduced, and we take a look at encoder/decoder schema.