The answers suggesting "Midi sequencer" are just wrong. A sequencer is for recording and replaying Midi signals but does not turn them into audio. In a similar vein, DAW or "Digital Audio Workstation" is wrong. While the scope of a DAW is ill-defined and may well come with a Midi softsynth, this is by no means a required component. Indeed, DAW is not even a guarantee for any Midi capability: the popular free DAW software "Ardour" gained some Midi recording capabilities only with version 3. However, pretty much every DAW these days can make use of Midi control surfaces. But those are for things like faders, rotary controls and similar, not musical instruments.
So after expounding on what you are not looking for (at least not primarily), the usual term for converting Midi to audio by software is "software synthesizer" or "Midi synthesizer" or even "realtime Midi player" where "realtime" sort-of suggests that the input to the player is not from a file but rather a Midi device.
In hardware, the term to look for is a "Midi expander".
Note that most of those terms apply for generic converters of Midi to audio. There are some software synthesizers specializing only on particular instruments, doing a lot of work possibly including physical modelling on them. Those can also be called "virtual instruments" and are loosely associated with VST plugins.
How many instruments and which kinds of instruments a softsynth supports is obviously an implementation choice, and even "full GM2 (General Midi 2) support" does not mean that all of the available instruments are useful in production. Particularly stuff like solo string instruments, flutes, and a few other instruments with an expressivity not easily measured in a single parameter (like the key velocity of a piano) are hard to get convincing: instrument sections are easier to get right.
For that reason, there is something to be said for just biting the bullet and getting a hardware Midi expander: something that's 20 years old still has been in the competitive music business for decades. Even if its modeling is not up to modern standards, its playability had to be. And stuff like "latency" had to be well-controlled even then. As opposed to a purely digital solution for song creation, however, you'll have to rerecord the audio signal and then need good quality audio inputs.
But at least the monitoring/auditory feedback will be immediate without configuration hassles, and that makes a difference in playability.