Audio: the sound has already been captured
An audio file represents the sound signal over time. You can cut it, crossfade it, process it with effects, change level, and—depending on the software—edit pitch or timing. The specific sound of the voice, microphone, guitar, or room is part of the material.
MIDI: the performance information stays flexible
MIDI does not store the finished sound. It can store which notes were played when and which controller values were used. That means the same part can later drive a different piano, synthesizer, or string sound without re-recording the notes.
A practical comparison
- A singer recorded through a microphone: audio.
- An electric guitar recorded as a signal: audio.
- A MIDI keyboard controlling a virtual piano: MIDI for the performance events; audio is produced during playback or rendering.
- A finished stereo mix: audio.
- Mod-wheel or sustain-controller movements: MIDI or related control data.
Both can live in the same project
A typical DAW session can contain audio and MIDI tracks at the same time. That is completely normal. MIDI might control a virtual drum instrument while vocals and guitars are recorded as audio. The audible results are combined into audio during the final mix or export.
They are not lossless translations of one another
Audio does not contain an unambiguous list of the keys that were originally pressed, and MIDI does not contain the individual microphone sound. Automatic audio-to-MIDI or MIDI-to-notation workflows can be useful, but they involve interpretation rather than a lossless format conversion.