The opening verse was cold, a beautiful automaton reciting its lines. Then, the silence. The tiny dip. Hana’s voice wavered, just for a frame of a second. And then she fell into the chorus. The growl on "yo-ake" was imperfect. It was ugly. It was real.
Kenji was tuning the voice of "Hana," a melancholic bank with a soft, breathy tone that cracked like autumn leaves. The song was his own—a desperate, quiet thing about a train station at 3 AM. He’d recorded a guide vocal, raw and flawed. His voice cracked on the bridge, right on the word "kaze" (wind). He wanted that crack. Not the perfect, AI-smoothed version of a crack, but that crack. The specific fracture of a specific human throat on a specific Tuesday night when the loneliness had felt like a physical weight. vocaloid 6 tuning
The old methods were still there, hidden under a drop-down called "Legacy Mode." He clicked it. The interface shifted, becoming the intimidating, spreadsheet-like nightmare of VOCALOID 3. Hundreds of dots. Envelopes for velocity, for pitch bend sensitivity. No AI to help him. Just him and the math. The opening verse was cold, a beautiful automaton
For the next three hours, Kenji became a micro-surgeon of silence. He inserted a tiny, 0.2-second dip in the Pitch Deviation right before the chorus—a moment of doubt, a slight downward glance before the leap upward. He manually painted a "Growl" parameter on the long, held note of "yo-ake" (dawn), not a full rasp, just a granular flutter, like sand slipping through fingers. He took the AI’s perfect, buttery portamento between two notes and replaced it with a jagged, stair-stepped curve, making Hana sound like she was choking on the word. Hana’s voice wavered, just for a frame of a second
The chorus needed lift. He selected the four bars and switched back to the AI "Dynamic Mode." He sang into his laptop’s cheap mic: "Kaze ga fuitara…" with a swelling, desperate rise in pitch. The AI parsed it. For a moment, Hana’s voice bloomed—rich, powerful, heartbreaking. But the transition from the flat, robotic verse to the AI-generated chorus was a cliff. A hard, digital step.
VOCALOid 6’s new "Expressive Control" feature was supposed to allow for this. It let you import an audio reference, and the AI would analyze the timbre, the portamento, the raw, ugly gasps for air. But when Kenji hit "apply," Hana’s voice emerged polished. The crack was there, but it was a diamond crack—symmetrical, beautiful, meaningless.
At 2:47 AM, he played it back.