MIDI Generation Tools

Getting MIDI rather than audio means every note is editable, and every sound is your problem.

Related

MIDI ProtocolVelocityDAW

The problem

Text to audio hands you a finished track, and changing one chord means starting over. MIDI generation is the opposite: you get notes that can be dragged, re-voiced and split across instruments. The price is that the sound is your job, and results often feel too correct, missing a performer's breath.

How it works

Symbolic models work on discrete events such as pitch, duration and velocity, conditioned on text, chord symbols or previous bars. Because audio is not involved, the model can concentrate on voice leading and rhythmic structure. But MIDI carries no timbre and no performance nuance, and quantised to a grid it easily sounds mechanical. Models also understand orchestration poorly, often writing melody and accompaniment at the same density.

How to use it

Treat it as a sketching assistant: generate a theme, a harmonic skeleton or a drum pattern, then edit. Fix velocity and microtiming afterwards so the part breathes. When changing instruments, reassess register and density instead of handing piano figuration to strings. Check generated parts for collisions. The MIDI can go into a DAW for further work or into notation software for engraving.

What to listen for

In the rondo, whether theme and accompaniment separate; in the quartet, whether four voices each speak; in the folk tune, whether ornaments feel natural. Judging MIDI means judging structure and performance, not tone.

Sources

• 符号音乐生成的表示方式，依公开研究摘要核对，未抄原文
• MIDI 生成与音频生成的取舍，依制作实践通识核对，未抄文本

Used for fact-checking only; all prose is original.