AI Transcription and Scoring
AI transcription turns a performance or audio into notation, with quantisation and phrasing as the main error sources.
Related
Turning a performance into notation means judging pitch, onset, metre and phrasing at once. Every judgement is ambiguous; a model simply makes the ambiguity visible.
The problem
Performances do not sit on the grid. Chords roll, and the pedal overlaps harmony. A transcription that tracks only pitch produces strange durations.
How it works
Audio transcription detects pitch and onsets, quantises to a grid, then assigns voices and rhythm. Polyphony and pedalling are hard because several notes sound at once and need not belong to one voice. Symbolic transcription infers metre, key and phrasing from MIDI. The model proposes; beats and voices still need confirmation.
Practice and judgement
Set metre and tempo before transcribing. Use a monophonic model first, then split voices. Check whether grace notes became full values, ties break across systems and rests are filled. Compare at slow playback, then export MusicXML and correct in notation software.
What to listen for
Hear whether accents match the performance, long notes split into several and pedalling muddies harmony. Then hear the rests; real breaths are filled with sustain.
Common misunderstandings
Transcription is copying a record; correct pitch is enough; automatic quantisation replaces beat judgement. Readability comes from phrasing, not note count.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Sources
- 自动音乐转写的音高、起音与量化流程，依公开论文核查，未抄文本
- 乐谱语义与 MusicXML 交换格式，依公开规范核查，未抄文本
Used for fact-checking only; all prose is original.