The process by which sounds cards produce audio output from MIDI input is called synthesis. There are three synthesis methods:
FM synthesis combines multiple sine waves of differing frequency and amplitude to produce a composite wave that resembles the native waveform of the instrument being synthesized. How close that resemblance is depends on the instrument and the quality of the FM synthesizer circuitry, and may vary from reasonably close to only a distant approximation. Even the best FM synthesis sound cards produce artificial-sounding audio, particularly for "difficult" instruments. Until the mid-1990s, most consumer-grade sound cards used FM synthesis, but FM synthesis is now obsolete.
Wavetable synthesis uses stored waveform audio samples of actual instrument sounds to reproduce music. The sample may be used as is, or modified algorithmically to provide a sound for which no sample is stored. For example, the wavetable may contain a stored sample of an actual violin playing an A note at 1760 Hz. If the MIDI score calls for a violin playing that A note, the sample is used directly. If the MIDI score calls for a violin playing an A note one octave higher (3520 Hz) and that note is not available as a stored sample, the synthesizer creates the 3520 Hz A note based on the data it has stored for the 1760 Hz A note. The quality of wavetable synthesis depends on the number, quality, recording frequency, and compression used for stored samples, and on the quality of the synthesizer hardware. Early wavetable sound cards were limited to the samples stored on their onboard ROM, which was typically 512 KB to 4 MB. Many current wavetable sound cards have, in addition to samples residing on onboard ROM, the ability to use additional samples stored on disk and loaded into main system memory as needed.
Although good wavetable sound cards produce music quality an order of magnitude better than FM synthesis sound cards, they are still limited by the finite number of stored samples, and so still must simulate much of their output instead of using samples directly. For example, whereas most wavetable sound cards store a total of perhaps 4 MB of samples for all instruments, fully sampling just a piano would require 10 MB or more of samples. That means that some piano sounds must be interpolated from existing samples, which in turn means that the piano emulation is not as good as it might be because interpolated sounds are less realistic than sampled sounds. The latest synthesis method, waveguide synthesis, also called physical modeling synthesis, creates a virtual model of an instrument and produces sounds based on calculations made against that model. Full waveguide synthesis for all instruments is beyond the hardware capabilities of any current PC. Some current high-end sound cards, however, use waveguide synthesis in conjunction with wavetable synthesis to improve the realism of emulation for some instruments.