Split two sounds into the same bank of frequency bands. Measure how loud the modulator (a voice) is in each band, and apply that moving loudness to the matching band of the carrier (a synth you play). The carrier ends up with the voice's spectral shape but the synth's pitch — so a sawtooth chord talks. No FFT here: it's a bank of band-pass filters plus an envelope follower per band (rectify, then smooth) driving a VCA — exactly the analog architecture. The built-in modulator is a synthesized voice sweeping through vowels, so it works with no mic; or drop a file, or speak in.
Fewer bands = more robotic; faster smoothing (higher Hz) tracks consonants but adds buzz.
hold a chord (a s d f …) and hear it speak — press ▶ to start the built-in voice