RoleChallenge-ranked multi-instrument AMT
Why it is usedUse as the MIROS comparison path: it ranked #1 overall in the 2025 AMT Challenge and kept a higher three-instrument F1 than YourMT3+, but it still degrades on dense polyphony and timbre confusion.
Output impactMulti-track MIDI
ModelYourMT3+ YPTF.MoE+Multi
RoleMulti-instrument MIDI transcription
Why it is usedStable TelkNet path with public checkpoint controls. The challenge paper still shows the hard limit clearly: YourMT3-YPTF-MoE-M drops from 0.7594 F1 on one-instrument excerpts to 0.3918 on three-instrument excerpts.
Output impactMulti-track MIDI draft
RoleHistorical baseline
Why it is usedPublic MT3-family benchmark baseline, shown for comparison only.
Output impactBaseline comparison only
RoleMuScriptor Multi-Instrument Transcription
Why it is usedMuScriptor-large leads YourMT3+ on the paper's shared private D_Test multitrack comparison and is designed for real-world mixed music.
Output impactOutputs Type-1 multi-track MIDI; 35 UI labels map to 36 learned groups. Velocity remains fixed at 100. Before conversion, choose a stabilized section-level adaptive tempo map (bar-aligned when meter evidence is reliable), one automatically detected BPM, or one manually entered BPM. After completion, playback speed and BPM can still be adjusted in the preview control, and downloads use its current setting.
ModelYourMT3+ YPTF.MoE+Multi noPS
RoleMulti-instrument MIDI transcription
Why it is usedStable TelkNet path with public checkpoint controls. The challenge paper still shows the hard limit clearly: YourMT3-YPTF-MoE-M drops from 0.7594 F1 on one-instrument excerpts to 0.3918 on three-instrument excerpts.
Output impactMulti-track MIDI draft