When sampling or recording audio, the audio wave is broken into sections or regions which are described by data bits .The number of these regions determines whether the sample is 8,16 or 24 bits. The greater the number of divisions, the closer the digital recording will be to the original sound. An 8 bit sample contains 256 steps of information whereas a 16 bit sample has 65,536 but a 24 bit sample has 16,777,216 .It is clear what word length is potentially closer to reality.
An area of concern is dynamic range ,which as we know determines our headroom and noise floor levels. The lower the noise floor and the higher the headroom, the better quality the audio will be.In terms of headroom every bit of data equates to 6 db of headroom with 8 bit having 48 db, 16 bit having 96db and 24 bit having 144db. Low bit rate samples even with smoothing filters tend to have a grainy, diffuse sound with poor clarity especially at high frequencies.
The next topic is sample rate or how many times per second are we going to measure our audio and assign a binary number to it. This parameter determines the frequency response or bandwidth of our audio stream with higher amounts of sampling providing a truer picture of reality. The Nyquist theorem states that a wave must be sampled twice to get a true representation of the wave. This is because the wave has both positive and negative peaks and each half must be recorded since they may be very asymmetrical.. Therefore if you sample at 44,100 times per second the highest frequency possible is 22,050 kHz,; 48,000 sampling frequency will yield 24,000hz max and of course 96khz sampling will yield 48,000hz max .