Weeks 5-6 · voice · cardinal vowel identity
Sing five vowels on one pitch. Each one keeps its identity.
The rule
pass: classifier confidence ≥ 0.60 on 4/5
Vowel space is a 2D map of mouth shape: tongue front/back on the horizontal, jaw close/open on the vertical. Each emerald ellipse is one of the five cardinal vowels. When you sustain /a/ the spectral centroid of your sound should land inside the /a/ ellipse — that’s pass. The grey dot stranded between /i/ and /e/ is the failure mode: a schwa-coloured production the classifier won’t commit to.
Hear it first
Listen to the five cardinal vowels sung on a single pitch, then the same five squeezed into a muddy “uh.” The contrast you’re training is between committed mouth shape and undifferentiated speech-default.
What’s happening.Each vowel is a unique resonance configuration of the vocal tract — characterised by two formant frequencies, F1 (controlled mostly by jaw opening) and F2 (controlled mostly by tongue position). Untrained singers often collapse all five vowels toward a neutral schwa under pitch pressure; trained singers keep F1 and F2 distinct so the lyric stays audible. Today we’re checking the second thing.
Why 2 seconds, not 4.The sustain needs to be long enough for the spectrum to settle (the first ~300 ms is unreliable) but short enough that this isn’t a breath endurance test — that’s a separate lesson. Two seconds gives the classifier 1.7 s of clean steady-state to decide on. The confidence gate of 0.60 is intentionally forgiving for week 5; future vowel work will tighten it.
Why this is the foundation.Every downstream voice lesson — vowel switching, vowel modification on ascent, consonant-vowel clarity, dynamics-vowel-color — assumes the classifier knows what your five vowels look like. If today’s production is muddy, those lessons can’t tell apart “wrong vowel” from “wrong everything.”
Session ahead
First-time setup: we’ll record your five vowels (~10 s total) before the session starts. Future lessons reuse the calibration.