Transkun open-source piano transcription model nails MIDI, sustain events included
alexanderchen · x · 2026-10-27
alexanderchen tried the open-source piano transcription model Transkun (Yujia-Yan/Transkun, 484 GitHub stars, MIT licensed) and reports it worked very well, including sustain events. Transkun is an audio-to-MIDI automatic piano transcription system built on neural semi-CRFs, with papers at ISMIR 2024 and NeurIPS 2021, installable via pip3 install transkun. He plans to combine it with Gemini 4 Argon to analyze and visualize his piano playing, tackling left/right hand separation from MIDI + video input.
Related event: Developer combines Transkun and Gemini 4 Argon to analyze piano playing(3 posts)→
More from Multimodal
- Single HTML/JS file generates creative art and copy in-browser with zero LLM calls — Afinetheorem · 2026-10-27
- $5k kale-img-1 fine-tuning contest won with MOSS-Image base, SDXL trails far behind — ostrisai · 2026-10-27
- FLUX.3 Edit nails a -90° car rotation on first try with just two reference images — EvilBot-666 · 2026-10-27
- Fan-Made AI Series 'Wolverine: Trail of Tears' Episode One Released — ArtificialLightPixel · 2026-10-27
- Seedance 2.0 Mini demo: stunning snow leopard video that can't actually roar — Burak_Antakya · 2026-10-27
- ComfyUI Krea2 Prompt Styler Node Revived in a Fork With a Massive Style Library — Winter_unmuted · 2026-10-27