Introduction to AI Music
AI music follows two paths, symbolic and audio, each solving a different generation problem.
Related
Making a machine produce music in a recognisable style is no longer hard. Making it follow structure, stay editable and carry clear rights is.
The problem
Music involves pitch, rhythm, harmony, orchestration and performance expression. Which layer is generated decides whether the output can be edited, how convincing it sounds and who owns it.
How it works
Symbolic systems treat notes and controllers as data, learn context at the event level and output MIDI or notation, where one note is easy to change. Audio systems generate waveforms or spectra, sounding complete but hard to take apart. Training compresses a corpus into statistical patterns; generation samples under a prompt and conditions for style, section and length.
Practice and judgement
Decide whether you need a demo, material or a finished track. Choose symbolic or layered output when edits matter, audio when the finished sound matters. Check key, section and duration controls, keep a human editing stage, and log tool versions and prompts for reproduction.
What to listen for
Hear whether section changes are abrupt, harmony sits in one place and repetition feels mechanical. Then hear tails and reverb, and ask whether the music ends or the file ends.
Common misunderstandings
AI composes; sounding similar is lawful; one generation is deliverable. Tools offer candidates; judgement remains human.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Sources
- 符号音乐生成与音频生成的综述结论，依 ISMIR 公开论文核查，未抄文本
- 生成式音乐工具的输入条件与输出形态，依公开文档核查，未抄文本
Used for fact-checking only; all prose is original.