When we speak into a microphone, these changes in pressure are converted to proportional variations in electrical voltage. Computers equipped with the proper hardware can convert the analog voltage variations into digital sound waveforms by a process called analog-to-digital conversion (ADC), which involves two separate components:

1. Sampling - Even though a waveform is usually depicted as a continuous function of time, Figure 2 below, a more detailed rendition of a few millseconds from the waveform of Figure 1 above, shows that the function is in fact discrete. Sampling means taking a fixed number of pressure value readings at equal time intervals from the continuously varying speech signal. For example, clean speech such as that depicted on this page is sampled 16000 times per second; its sampling frequency is 16 kHz. If you count carefully, you will find a total of sixteen dots per millisecond in Figure 2. Telephone speech is sampled at half that rate - 8000 kHz. On the other hand, compact disk recordings have a sampling rate of 44.1 kHz. The higher the sampling rate, the better the sound quality, but the more bits required.

2. Quantization - Each sampled pressure value is rounded or quantized to the nearest value which is expressible in a given number of bits. There is a direct relationship between the accuracy of quantization and the number of bits required. Clean speech such as that depicted here often uses 16 bits, for a total of 65536 possible quantization levels, while telephone speech is accommodated in 8 bits for a total of 256 quantization levels.