Open-source vocal synth models vocal tract as acoustic tube network, adds MusicXML support
TinfoilTricorn · x · 2026-11-08
Developer jamesonhodge shared an open-source vocal synthesis project that, unlike Vocaloid's approach of stitching and pitch-shifting phoneme recordings, models the vocal tract as a network of branching acoustic tubes. Naturalness is still being improved by tuning against recordings of real singers, but early results already sound impressive. Meanwhile, community member yungalgorithm opened PR #24 to convert MusicXML notation into notes, enabling fun arrangements and performances from MusicXML files.
Related event: Developer Shows Singing Synthesis via Branching Vocal Tract Network(2 posts)→
More from Multimodal
- Blender user cracks 360-degree video tracking, demo looks impressive — PurzBeats · 2026-11-08
- Redditor asks how to make clean 2D animated reels with Claude — Illustrious-Low-1123 · 2026-11-08
- Gemini animation trick: one image in, layered crash-zoom animation out — jnack · 2026-11-08
- ComfyUI workflow for MiniMax H3 character swap: 64s 1080p video in 15 segments — wjc_5 · 2026-11-08
- Full GPT Image 2 prompt template for pastel paper-collage city posters — SimplyAnnisa · 2026-11-08
- Cat Steals Package, Two Drones Give Chase in Seedance 2.5 AI Short — nikola_mr64990 · 2026-11-08