for harpsichord, two voices — composed by Sérenda
Sérenda is a system that composes baroque music. It mines a corpus of Bach for structure, generates the piece under a counterpoint solver, and renders it through a sampled consort in Werckmeister III at A4 = 415 Hz.
Streams as Ogg Vorbis (2.1 MB). The lossless FLAC is linked below.
I had ChatGPT mark it blind, on the reasoning that asking the system that helped build Sérenda to grade Sérenda's output is worthless. It returned 25/30, which it described as solid Grade 8 technique — and one criticism I have not been able to unsee:
The architecture is also highly visible. I can frequently predict what the machine is going to do next: entry, two-bar episode, entry, two-bar episode… The music therefore sometimes feels like an extremely good demonstration of a contrapuntal system rather than an entirely convincing piece of human musical rhetoric.
That is one language model's judgment using ABRSM bands as borrowed vocabulary — not a human examiner, and not an official mark. The critique is the useful part, and it is what the current work is aimed at: the engine does not need more machinery, it needs the machinery to become less visible.
I aim to publish Sérenda, open source on GitHub in December 2026.