Last updated February 10, 2015 08:39, by Herv√© Bitteur

Handling examples from Recordare with Audiveris

This section currently applies to Audiveris V4.x, it is to be updated for V5.

Nota: Among Audiveris distribution examples, you can find two examples originating from Recordare by courtesy of Michael Good. They are named Dichterliebe01.pdf and SchAvMaSample.pdf respectively. If you are interested in the other Recordare examples, you can download all the examples set by yourself directly from MakeMusic site

These sometimes complex examples are used as representative tests for Audiveris. The goal for the current release of Audiveris was to be able to completely process each of these examples, even if some manual corrections were still needed.

Nota: We assume that you have chosen the proper OCR language before launching the transcription of the score at hand. Even in this case, you cannot expect that all words will be correctly OCR'ed.

For each example, the sections below describe:

  1. the specificities found, using the precise pixel location in page whenever possible
  2. additional information, such as the manual correction involved



Special Items Comment
This is really an awfully challenging score in terms of technical difficulties and of computing resources (memory and CPU) Setting the application memory to at least 1 GB is recommended. Since we have just one system per page and this system is really huge, rebuilding the system on each user modification takes a few seconds.
Advice: You'd better disable the auto-rebuild (in Score pull-down menu) when editing the score, and manually launch the rebuild when you are through with the modifications.
The score exhibits advanced part organization with the notion of "part-group" which Audiveris does not handle yet. Audiveris supports BRACKET or BRACE for a part but not a combination of them. So, while this is not perfect, select the shape you prefer for each part individually
Many long horizontal slurs are broken due to the segmentation brought by staff line removal Select the slur pieces and assign the SLUR shape to the compound
Similarly several void node heads are stuck to slur pieces, resulting in wrong size and location Deassign these false symbols, until just the note head is recognized as such, besides this will fix the slots locations errors



Special Items Comment
pixel(1830,1879): Slashed grace note needs to be translated Grace notes, as well as cue notes, are not yet implemented in Audiveris
pixel(1998,2022): Word "espressivo" not OCR'ed Assign manually the TEXT shape to the glyph
pixel(2312,1990): Word "dim." not OCR'ed Assign manually the TEXT shape to the glyph
pixel(1871,1852): Note inconsistency in chord A small slur has been mistaken for a void note head.
Manually deassign the glyph.
pixel(692,1937): Note inconsistency in chord. We have a stack of 3 heads, one black and 2 void. The black head glyph, attached to the right stem, goes too far below and gets connected to the left stem. Use section selection mode to select just the sections that pertain to the upper note, and assign the NOTEHEAD_BLACK shape to the resulting glyph
pixel(2243,2132): Dot unassigned The augmentation dot is too far down and could not be connected to a note head. This is harmless since the chord already has a dot attached.



Special Items Comment
pixel(375,2371): Invalid dynamics signature. The word "hin," has not been correctly OCR'ed, select the word and assign the TEXT shape to it
pixel(1161,759): A large slur is broken into two smaller ones. Select both glyphs and assign SLUR shape to the compound.
pixel(1879,2499): A slur is stuck to an eighth glyph. Use section mode and select the sections that compose the eighth symbol. Assign the EIGHTH shape to it.
pixel(646,988): A piano sign (p) has been mistaken for a word. Assign the DYNAMICS_P shape to it
pixel(1310,730): how to manually enter the "extension" sign (something like a '_') directly in the GlyphBoard? Not yet implemented



Special Items Comment
The notation used is a tablature Audiveris does not handle tablatures for the time being



Special Items Comment
pixel(2101,592): Direction "dolce e leggiero" not OCR'ed Use a lasso to assign the TEXT shape to the glyph.
pixel(540,951) Voice #1 too long A rather vertical slur has been mistaken for a QUARTER_REST. Assign it the SLUR shape
It will complain that the slur embraces no note, because grace notes are not yet implemented
pixel(1469,2441): Direction "dim." not OCR'ed Manually assign the TEXT shape to the glyph
pixel(1444,2845): Another direction "dim." not OCR'ed Manually assign the TEXT shape to the glyph
pixel(1408,2455): A deassigned slur Manually assign the SLUR shape to it
pixel(1164,2737): A pack of 2 note heads plus a flag is mistaken for FLAG_3_UP Manually deassign the shape, it will be correctly recognized as 2 heads + 1 flag



Special Items Comment
No specific error on page #1
Errors below relate to page #2
pixel(318,1267): Could not left-connect slur This is due to unrecognized slur at end of previous system at location (2371,529)
pixel(831,1451): A flag, stuck to a slur chunk, is mistaken for a NOTEHEAD_VOID Manually deassign (twice) the glyph until it gets recognized correctly
pixel(2335,1121): Lyrics word "da" not OCR'ed Manually assign the TEXT shape
pixel(1832,1521): A glyph composed of half a slur, a note head and a flag is mistaken for a NOTEHEAD_VOID Manually deassign this glyph (several times) until it gets correctly recognized
pixel(2038,1998): Dot unassigned The slur is broken into pieces. Select all pieces with a lasso, and assign the SLUR shape
pixel(1142,2796): Direction "ritard." not OCR'ed Manually assign the TEXT shape
pixel(655,2743): Glyph composed of 2 beams not recognized Manually assign the BEAM_2 shape
pixel(154,950): Sentence with no role defined This occurs frequently with initial measure numbers at the beginning of a system. To get these numbers exported, you can assign them the "Direction" role.



Special Items Comment
"Smart" inter-system borders are not correctly computed Switch to a boundary modification session to manually adjust the broken lines. When done, end the boundary session so that the modifications are taken into account
Several '3' tuplets are not recognized Manually assign the TUPLET_3 to them, this will fix the "Voice too long" error messages
pixel(982,1886): a tuplet with brackets (+- 3 -+) was mistaken for text. Deassign the glyph and manually assign the TUPLET_3 shape to the '3' digit. Audiveris does not handle tuplet brackets for the time being.
pixel(1365,2324): Note inconsistency in chord We have a chord with both a black head at the top and a void head lower on the stem. The upper (void) head was mistaken for a black head because it was part of a two-head glyph, incorrectly assigned the NOTEHEAD_BLACK_2 shape. The solution is to separate the two heads, by switching to section selection mode, collecting the sections that pertain to the void head only and assign the proposed NOTEHEAD_VOID shape. The other head will automatically be recognized as a NOTEHEAD_BLACK.



Special Items Comment
In first system, the lyrics "How vain___ is man___ who ..." is not OCR'ed. Assigning the TEXT shape manually to the whole lyrics line does not work either. This OCR problem seems due to the presence of "___" characters in the line. Assigning the TEXT shape to the words by carefully avoiding to select the "___" characters works correctly though.
pixel(312,599): The COMMON_TIME signature has been assigned a 3/4 value Though the "C"-shape glyph was recognized with a confidence of 99.8%, the time signature inferred from the measures in this page was found as 3/4, and used to overwrite the time signature.
Manually reassign the COMMON_TIME shape to the glyph and all parallel time signatures will get corrected at once.
pixel(1811,1658): A large glyph composed of a sharp and a natural stuck together Use section mode to select the sections pertaining to sharp, and assign the proposed SHARP shape.
Do the same for the natural sign.
pixel(1511,642): Unrecognized double beam Manually assign the BEAM_2 shape to the glyph



Special Items Comment
pixel(520,466): The "Alla Marcia" direction was not correctly OCR'ed Manually assign the TEXT shape, it will get OCR'ed correctly and its direction role detected.
pixel(1905,1409): In this measure, there are 6 note black heads of rather small size, not recognized as such Manually assign the NOTEHEAD_BLACK shape of each of them
pixel(298,1574): The word "schickt" was not OCR'ed Manually assign the TEXT shape, it will be OCR'ed and its lyrics role recognized
pixel(2360,1405): The staff, as well as the other staves in this system, finishes with a 5/4 time signature which is mistaken for a clef These ending time signatures don't matter when the score is exported. If you wish, you can manually deassign them or even assign them the right shape.



Special Items Comment
pixel(1120,535) and pixel(1499,533): Under the TURN signs these are a small natural and a small sharp sign, unrecognized You can manually assign the proper shape to them. However, how should they be translated into the exported score data? (to be investigated)
pixel(1828,1191): The lyrics word "en," has not been OCR'ed Manually assign the TEXT shape. If the text content is still blank, manually type the content "en," in the glyph board. The lyrics role will be recognized at once.
pixel(2194,1171): The lyrics word "len," has not been OCR'ed Manually assign the TEXT shape.



Special Items Comment
Many GRACE_NOTE_NOSLASH glyphs are not translated Though the shape is often recognized, the glyph is not yet translated (this feature is not yet implemented)
pixel(1015,796): Dot unassigned Actually it's a small slur linked to a grace note. Since grace notes are not yet handled, the best choice is to deassign this glyph
pixel(1846,1077): Similarly, this slur glyph is mistaken for a dot and even assigned the AUGMENTATION_DOT shape, which leads to a too long voice in the containing measure Simply deassign the glyph
pixel(350,2682): A double beam has not been recognized Manually assign the BEAM_2 shape



Special Items Comment
pixel(1560,2444): The detection of border between systems 3 & 4 failed at this point. Enter a boundary edition and modify the broken line so that it gets above the large slur at pixel(1803,2411) and below the sharp sign at pixel(1821,2379). You have to be very precise!
pixel(1125,569): This slur has been mistaken for an OTTAVA_ALTA. Manually deassign the glyph, it should get the SLUR shape by itself.
pixel(494,966): The "col Pedale" direction statement has no recognized role Manually assign the Direction role
pixel(663,555), pixel(874,551), pixel(1069,765) & pixel(1305,743): The four "6" signs have been mistaken for breath marks or characters or not recognized at all Manually assign the TUPLET_6 shape to each of them and the "Voice too long" error will go away.
Except in first measure, the 6-tuplets are not indicated by proper TUPLET_6 sign. The score assumes the reader will implicitly infer the TUPLET_6 for the following measures, something Audiveris does not do yet.
A tedious workaround is to manually insert (via drag n' drop from the Shape palette) artificial TUPLET_6 signs at proper locations.
pixel(1436,245) Dot unassigned Assign the whole title "Ave Maria", including the dot, to the TEXT shape, it will be correctly OCR'ed this time.
pixel(396,1202): Similarly the whole set of 3 lyrics lines in system #2 is not correctly OCR'ed Select the 3 lines as a whole and assign the TEXT shape to the compound
pixel(2130,1743) and pixel(2296,1744): Two tuplets 3 have not been recognized. Manually assign the TUPLET_3 shape to each of them
pixel(2083,1033) & pixel(1815,2532): Beam with no chord on (left/right) stem on these locations. It is actually a beam between cue notes, something not yet implemented by Audiveris. Simply deassign the beam glyph for now (perhaps several times)
pixel(663,2530) & pixel(1186,2530): Double beams not recognized Manually assign the BEAM_2 shape at these locations
  • Mysql
  • Glassfish
  • Jruby
  • Rails
  • Nblogo
Terms of Use; Privacy Policy;
© 2014, Oracle Corporation and/or its affiliates
(revision 20160708.bf2ac18)
Please Confirm