Q & A(I)

We asked for your questions, and you sent so many that one episode couldn’t hold them. This is part one of two — and part one is brought to you by two letters: A and I. That wasn’t the plan; it’s just where the mailbag pointed. Philip Rothman and David MacDonald take on the growing expectation that artificial intelligence should be able to turn a scan, recording, or generated track into finished, usable sheet music. Why is that still so difficult? And where can AI genuinely help musicians and music preparation professionals today? The discussion ranges from optical music recognition and audio transcription to proofreading, house styles, sample libraries, score following, and the potential for language models to communicate directly with notation software. Along the way, Philip proposes the “Gould test” as the music preparation equivalent of the Turing test. Underneath it all are larger questions about the value of expertise, the role of human judgment, and where responsibility lies when the technology gets it wrong. The rest of the mailbag — the questions that have nothing to do with AI — is coming next month. Products mentioned Notation software Dorico (Steinberg) Sibelius (Avid) MuseScore Studio (Muse Group) Flat (Tutteo) StaffPad Scanning and optical music recognition NoteVision (Muse Group) Soundslice sheet music scanner Newzik PlayScore 2 SmartScore 64 Pro (Musitek) Opuscan (Tutteo) Audiveris (open source) Audio transcription Klang.io and Transcription Studio Songscription Basic Pitch (Spotify) (open source) Magenta MT3 (Google) (research) Playback and vocal synthesis NotePerformer (Wallander Instruments) Cantai (Turing Opera Workshop) AI arranging and orchestration ArrangementLabs NotationAI ScoreSynth Orchidea (IRCAM / HEM / UC Berkeley) Score following and page turning Tomplay Smart Page Turn Smartleggio Piascore SampleSumo Music Following Antescofo (IRCAM) MCP bridges (community projects) mcp-musescore dorico-mcp-server (also on