Audiveris – 开源光学乐谱识别软件推荐
Audiveris 是一款专为音乐爱好者和专业人士设计的光学乐谱识别 (OMR) 软件。它的目标是将乐谱图像转录为符号形式，从而实现音乐的播放、编辑、搜索和再发布等功能。Audiveris 集成了一个高效的 OMR 引擎和用户友好的 OMR 编辑器，提供了良好的识别性能，尤其对于现实世界中的乐谱。
译者注：原文为 Optical Music Recognition，字面为「光学音乐识别」。为方便理解，译为「光学乐谱识别」。
OMR 与 OCR 的区别
OMR：光学乐谱识别
OCR：光学字符识别
在讨论光学乐谱识别时，常常会提到光学字符识别 (OCR)，但它们之间有几点重要区别：
特征化书写系统：音乐记谱是一种特征化的书写系统，包含丰富的视觉元素（如音符、附点、音符杆等），而字符识别则主要关注已定义的字母和单词。
语义恢复：OCR 只能对字母和单词的字形进行识别，而 OMR 不仅要识别乐谱，而且要恢复其语义，例如通过音符的垂直位置来翻译成音高。这种复杂性在字符识别中是没有相应的。打个比方，从乐谱图像中恢复音乐可能与从网站屏幕截图中恢复 HTML 源代码一样具有挑战性。
字符集的复杂性：尽管不像汉字等书写系统字符集数量上的宏大，OMR 的原始符号集在尺寸上变化更为广泛，从微小的元素（如音点）到可能覆盖整页的元素（如大括号），并且某些符号如连音线没有严格的定义，呈现形式也不一而足。
二维空间关系：音乐符号的空间关系是二维的，而文本识别通常是一维的流式信息，只需确定基线即可读取。
主要特点：
良好的识别效率，适用于现实中的乐谱（例如 IMSLP 上的乐谱）。
支持大规模乐谱处理，最多可达数百页。
方便的用户界面，能够检测和纠正大部分 OMR 错误。
兼容 Windows、Linux 和 MacOS 平台。
开源软件，数据透明。
Audiveris – Open Source Optical Music Recognition Software Recommendation
Audiveris is an optical music recognition (OMR) software designed for music lovers and professionals. Its goal is to transcribe music score images into symbolic form, so as to achieve functions such as music playback, editing, search and redistribution. Audiveris integrates an efficient OMR engine and a user-friendly OMR editor, providing good recognition performance, especially for real-world music scores.
Translator's Note: The original text is Optical Music Recognition, which literally means "optical music recognition". For ease of understanding, it is translated as "optical music recognition".
Difference between OMR and OCR
OMR: Optical Music Recognition
OCR: Optical Character Recognition
Optical Character Recognition (OCR) is often mentioned when discussing optical music recognition, but there are several important differences between them:
Featured Writing System: Music notation is a characterized writing system with rich visual elements (such as notes, dots, note stems, etc.), while character recognition focuses on defined letters and words.
Semantic recovery: OCR can only recognize the glyphs of letters and words, while OMR not only recognizes the musical notation but also recovers its semantics, such as the vertical position of the notes translated into pitch. This complexity has no counterpart in character recognition. For example, recovering music from a musical notation image can be as challenging as recovering HTML source code from a website screenshot.
Complexity of character set: Although not as large as the character sets of writing systems such as Chinese characters, the original symbol set of OMR varies more widely in size, from tiny elements (such as dot) to elements that may cover the entire page (such as curly brackets), and some symbols such as slurs are not strictly defined and have different presentation forms.
Two-dimensional spatial relationship: The spatial relationship of music symbols is two-dimensional, while text recognition is usually one-dimensional streaming information that can be read by simply determining the baseline.
Main features:
Good recognition efficiency, suitable for real-world musical notations (such as those on IMSLP).
Supports large-scale musical notation processing, up to hundreds of pages.
Convenient user interface capable of detecting and correcting most OMR errors.
Compatible with Windows, Linux and MacOS platforms.
Open source software with data transparency. |
