arXiv:2511.09562cs.SD2025-11

可视化多个MIDI音轨对比,辅助音乐转录模型评估。

WaveRoll: JavaScript Library for Comparative MIDI Piano-Roll Visualization

  • 基于浏览器的交互式多轨音轨对齐显示
  • 支持时间对齐与同步播放,精准定位音高/时序差异
  • 适合音乐生成与转录模型的误差分析与调试

WaveRoll 是一个交互式 JavaScript 库,可在浏览器中实现多个 MIDI 音轨的对比可视化与同步播放。它针对自动音乐转录(AMT)中的评估需求,用于对比同一输入产生的多个 MIDI 输出。该库在单一时间对齐的网格上展示多条音轨,并同步播放音频,使用户能够比较音高与时序差异,识别漏音或多余音符,观察起始与终止时刻偏差,以及段落级模式。我们预期此类对比有助于模型评估与错误分析,提升对模型行为的理解。该开源库已发布于 https://github.com/crescent-stdio/wave-roll。

原文摘要 · Abstract (English)

WaveRoll is an interactive JavaScript library that enables comparative visualization and synchronized playback of multiple MIDI piano rolls on a browser. It addresses a specific evaluation need in Automatic Music Transcription (AMT), contrasting multiple MIDI outputs produced from the same input. The library displays multiple MIDI tracks on a single, time-aligned grid with synchronized audio, allowing users to compare pitch and timing, identify missed or extra notes, and observe onset and offset differences, as well as section-level patterns. We expect that such comparisons would assist in model evaluation and error analysis, and help readers to understand the model behavior better. The open-source library is available at https://github.com/crescent-stdio/wave-roll

音乐生成MIDI可视化模型评估

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。