recomend ac i on

28
MUSIC HACK DAY, Vienna 2013 Audiveris Optical Music Recognition Presented by Hervé Bitteur [email protected] February 2, 2013 V1

Upload: vlayvladyypratonsconsky

Post on 13-Dec-2015

226 views

Category:

Documents


2 download

DESCRIPTION

recomendaciones antes de abrir guitar pro

TRANSCRIPT

Page 1: Recomend Ac i On

MUSIC HACK DAY, Vienna 2013

AudiverisOptical Music Recognition

Presented by Hervé [email protected] February 2, 2013

V1

Page 2: Recomend Ac i On

2www.audiveris.orgVIENNA 2013

Agenda

● User demo● The big picture● Audiveris technology● Current 4.2 release● Integration● Future

Page 3: Recomend Ac i On

3www.audiveris.orgVIENNA 2013

User Demo

● User oriented Input Corrections Outputs

● Case Input : Dichterliebe01.pdf Language : DEU Tempo : 80 Output : Finale Notepad

Page 4: Recomend Ac i On

4www.audiveris.orgVIENNA 2013

Score Transcription « big picture »

SymbolicNotation

PrintedScores

OMR

MusicInfo

Retrieval

IndexSearch

Edition

ScoresImages

EditViewPlay

Transpose

Pivot Format

Optical Music Recognition

AudiverisAudiveris

MuseScore

PeachNote

Scan

MusicXML

Page 5: Recomend Ac i On

5www.audiveris.orgVIENNA 2013

Music notation

● MusicXML Symbol level Meant for score interchange

● MIDI Note level Meant for digital instruments

● MP3 Sound level Meant for recorded music

High level

Low level

Page 6: Recomend Ac i On

6www.audiveris.orgVIENNA 2013

OCR vs OMR

OCR(text)

OMR(music)

Complexity Horizontal lines of characters

Combined horizontal & vertical directions

Technology Rather mature Several yearsbehind OCR

Products Many Just a few. One FOSS: Audiveris

Page 7: Recomend Ac i On

7www.audiveris.orgVIENNA 2013

Basic definitions: Runs & Sections

Black run of 3 pixels

White run of 18 pixels

Section of 6 adjacent runs

Page 8: Recomend Ac i On

8www.audiveris.orgVIENNA 2013

Tech Demo

● Technology oriented Runs Grid

● Case Input : Chula.png Language : FRA Params : showRuns, showCombs Output : MuseScore

Page 9: Recomend Ac i On

9www.audiveris.orgVIENNA 2013

Pixels: Staff Lines & other objects

Page 10: Recomend Ac i On

10www.audiveris.orgVIENNA 2013

Black runs histo. → line thickness

Page 11: Recomend Ac i On

11www.audiveris.orgVIENNA 2013

White runs histogram → interline

Page 12: Recomend Ac i On

12www.audiveris.orgVIENNA 2013

Vertical runs > line thickness

Page 13: Recomend Ac i On

13www.audiveris.orgVIENNA 2013

Vertical runs <= line thickness

Page 14: Recomend Ac i On

14www.audiveris.orgVIENNA 2013

Short horizontal runs put aside

Page 15: Recomend Ac i On

15www.audiveris.orgVIENNA 2013

Long hori. runs → lines skeleton

Page 16: Recomend Ac i On

16www.audiveris.orgVIENNA 2013

Clusters of equidistant lines

Page 17: Recomend Ac i On

17www.audiveris.orgVIENNA 2013

Staff lines detected

Page 18: Recomend Ac i On

18www.audiveris.orgVIENNA 2013

Staff lines removed

Page 19: Recomend Ac i On

19www.audiveris.orgVIENNA 2013

Wavy staff lines: Splines

Page 20: Recomend Ac i On

20www.audiveris.orgVIENNA 2013

Wavy staff lines: Dewarped image

Page 21: Recomend Ac i On

21www.audiveris.orgVIENNA 2013

Stems detected

Page 22: Recomend Ac i On

22www.audiveris.orgVIENNA 2013

Sections → Glyphs → Shapes

Page 23: Recomend Ac i On

23www.audiveris.orgVIENNA 2013

Generated score

Page 24: Recomend Ac i On

24www.audiveris.orgVIENNA 2013

Audiveris version 4.2● Released in December 2012● Features:

Ghostscript integration– for PDF input

Adaptive Binarization– for non-uniform images

Tesseract OCR V3– for text handling

ART-based moments– for shape recognition

Plugins– for MusicXML consumers

Page 25: Recomend Ac i On

25www.audiveris.orgVIENNA 2013

Local OMR Integration

MusicXML

scan

Audiverisplugin

plugin

MuseScore

display

client

Page 26: Recomend Ac i On

26www.audiveris.orgVIENNA 2013

OMR as a Service (hack of the day)

MusicXML

scan

MuseScore

display

internet

Audiveris

FrontEnd

client server

Page 27: Recomend Ac i On

27www.audiveris.orgVIENNA 2013

Audiveris evolution

● Technology Good building blocks available 4.2+: iterative approach OK for good-quality scans 5.x: fuzzy logic needed for poor-quality scans

● Architecture Focus on OMR service

● « Freemium » model Free for basic features or low volume Premium for more serious usage

Page 28: Recomend Ac i On

Thank you

www.audiveris.org

Audiveris [latin] := « you will have heard »

[email protected]