Why Harmony Is Harder Than Melody
Blob Opera combines four represented voice types in one responsive ensemble. Google says the team created two models: one taught the computer to sing and another taught it to harmonize. The official experiment page describes the second model as letting the blobs respond to and harmonize a user's input in real time. It does not publish a rule set or training corpus for that model.
How the Engine Monitors Your Input
When you move a blob, the remaining voices react around that input. This is why one gesture can sound like a coordinated group rather than four unrelated tones. Describing the audible result does not tell us how often the system updates, whether it predicts future chords, or which musical features it evaluates internally; those details are not in Google's public account.
Voice Leading Principles
Voice leading is the study of how individual musical lines move while forming harmony together. Small, singable movements are common, but real music uses many styles and exceptions. Blob Opera can prompt a listener to notice these relationships. Google has not stated that its harmony model performs a rule-based search or follows a specific textbook treatment of four-part writing.
Why It Sounds Musical, Not Mechanical
The experience feels collaborative because the user's gesture and the model's response are heard together. Listeners may describe some results as expected, surprising, stable, or tense, but those impressions should not be turned into undocumented claims about controlled randomness, delayed entries, or the model selecting a particular chord type.