引导式测验 · 中等 等级

音频测验中的起始检测

了解音频开始检测器如何发现事件攻击、使用频谱通量和峰值拾取以及处理渐进或重叠的声音。

相关引导路径音频中的起始检测
问题 1 的 8

音频中的“开始”到底是什么？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 音频中的“开始”到底是什么？
    • 声学事件开始的那一刻，例如音符的起奏
    • 声音的淡出
    • 曲目的平均响度
    • 一首歌的结尾
  2. What does spectral flux measure for onset detection?
    • A rule that marks every beat as a note onset
    • Positive increases in spectral energy between successive frames
    • A decay detector that selects only falling energy
    • A pitch estimator that finds the nearest MIDI note
  3. 为什么在光谱通量中应用半波整流？
    • 放大衰变
    • 将音频转换为文本
    • 因此，只有上升（正）能量变化才有贡献，忽略衰变
    • 减慢音频速度
  4. 哪种类型的发作最难检测？
    • 尖锐的鼓声敲响
    • 一声响亮的钹声
    • 军鼓后拍
    • 柔和、渐进的小提琴声
  5. 具有自适应阈值的峰值拾取阶段可以防止什么？
    • 双触发和假启动
    • 音频压缩
    • 变调
    • 立体声扩宽
  6. What can a neural onset detector learn from labeled audio examples?
    • A guarantee that every onset will be detected
    • Patterns in the audio features that signal event beginnings
    • A fixed tempo that applies to every recording
    • The identity of every instrument without evaluation
  7. What does onset backtracking do when detected times are used to create audio slices?
    • Moves each onset to the end of the entire recording
    • Replaces each onset with the nearest musical beat
    • Moves an onset to the nearest preceding local minimum of an energy signal
    • Adds a fixed one-second delay after every peak
  8. What can happen when an onset model is transferred to a different instrument or dataset?
    • Its predictions are guaranteed to remain identical
    • The audio automatically becomes monophonic
    • The model no longer needs a time-frequency input
    • Its performance may change and should be evaluated in the new domain