Fonts

since 5.3

Audiveris uses specific fonts for music symbols and for text symbols.

Table of contents

  1. Music fonts
    1. Head symbols
    2. Other symbols
    3. Music fonts hierarchy
  2. Text fonts
  3. Selection of fonts families

Music fonts

Before 5.3 release, only one music font family was used (Musical Symbols font initially, later replaced by Bravura font for SMuFL compliance).

Since 5.3 release, we can tell Audiveris which music font family to use by default, or for a given book or sheet.

Head symbols

This choice is key for head symbols, since the OMR engine uses a template matching technique to detect them during the HEADS step.

Here below are samples of just two significant head shapes – black head and cross head – for each of the available music font families (Bravura (the default), Leland, Finale Jazz and Jazz Perc).
Pictures are magnified 4 times for a better reading:

Shape / Font Family Bravura Leland Finale Jazz Jazz Perc
Black Head
and
Cross Head

The head templates are built upon a specific font:

  • its family is the user-selected music font family,
  • its precise size is derived from the measured staff interline.

Due to the way template matching works, it is essential to carefully check the match between the input score image and the selected font family.

  • The major difference is between standard fonts (Bravura and Leland) and jazz fonts (Finale Jazz and Jazz Perc), especially for all the cross-like shapes.
  • Within the jazz fonts, Finale Jazz and Jazz Perc differ mainly by their width, and the impact on template matching is more visible on oval-like shapes.

Other symbols

For symbols other than head symbols, such as Clef, Flag, etc, recognition is not based on template matching but on a neural network.

With proper training on representative samples, the neural network is able to recognize their shape, regardless of the selected music font family.
We can observe its top 5 results in the Glyph Classifier board.

At this point, the reader may wonder why this neural network approach is not also used for head recognition… The reason is the current network needs a glyph to work upon, and there is yet no way to isolate a head glyph reliably. 1

Music fonts hierarchy

Not all symbols are available in every music font family.

To cope with this situation, Audiveris has defined a hierarchy between music font families. Hence, when a given symbol is not available in a family, the families above in hierarchy are transitively searched until the symbol is found.

The current family hierarchy is as follows:2

    Bravura
    ├── Leland
    ├── Finale Jazz
    │   └── Jazz Perc
    └── Musical Symbols

Text fonts

Texts recognition is less sensitive to font family, compared to music symbols recognition.
The main reason is that Audiveris delegates texts recognition to Tesseract OCR.3

Since 5.3 release, we can tell Audiveris which text font family to use among the families available (Sans Serif (the default), Serif and Finale Jazz Text):

Sample / Font Family Sans Serif Serif Finale Jazz Text
Ophelia

A proper choice of text font family will result in a better consistency between the input image and its displayed transcription.

Selection of fonts families

In interactive mode, we can use the Book → Set book parameters dialog in its Music font and Text font fields:

In batch mode, we can make choices on the command line interface.

  • For music, choice is between Bravura, Leland, FinaleJazz and JazzPerc: 
    -constant org.audiveris.omr.ui.symbol.MusicFont.defaultMusicFamily=JazzPerc
  • For text, choice is between SansSerif, Serif and FinaleJazzText: 
    -constant org.audiveris.omr.ui.symbol.TextFont.defaultTextFamily=FinaleJazzText

  1. There are on-going works on global recognition without prior glyph segmentation. See this Audiveris Wiki article in its “6.x prototype” section. ↩

  2. The old “Musical Symbols” family has been totally replaced by “Bravura” family. It is kept only for samples generation from old .omr projects, something the end user can safely ignore. ↩

  3. In some cases, we have observed rather poor OCR results with a Jazz-like text font, perhaps because lowercase characters are displayed as small uppercase characters. To be further investigated. ↩