Peachnote Massive OMR recognized 1,6 M music sheets, 500 M notes - PDF document

Vladimir Viro Sponsored by Ludwig-Maximilians-Universität Munich vladimir@viro.name Peachnote Massive OMR recognized 1,6 M music sheets, 500 M notes multiple collections: IMSLP, Library of Congress, Duke, more to come poly-OMR: virtualized workflow supports all publicly available OMR packages Technologies Hadoop and HBase for scalable storage, computation and serving VMware and AutoIt for running OMR software Google App Engine and GWT for security and scalability of the frontend Google Analytics for collecting and querying usage data Music Ngram Viewer and Search Engine peachnote.com Search for scores containing melodies and chord sequences with or without rhythm Simple inverted index with billions of entries, but highly compressed(!) thanks to HBase --- 69 5 -12_5_4 -12_3 13 0 4 -9:1870 : 1 69 5 -15 3 7 3 -13 3:1870 : 23 69 5 -19 9 -4_11 -16_7:1871 : 2 --- Different query entry modes Usage Melody search engine at IMSLP 500-1000 daily visitors from over 170 countries More than 300 users visited more than 100(!) times each Usage data used to improve search results Sharing Have an algorithm working with symbolic music data? API All functionality and data Let it shine on the largest symbolic music data set! exposed to the world N-gram data released See a way of enriching your application with under CC-By License scanned sheet music or statistics derived from it? Please get in touch! http://www.peachnote.com/{api,datasets}.html

Peachnote ● Melody search for digitized music scores ● Music N-Gram Viewer ● Embeddable Score Viewer ● Collaborative score annotation system ● 170,000 users since launch of the N-Gram Viewer ● 1,000,000 users from 200 countries a month for the new Score Viewer and annotation platform

N-Gram Viewer and Search Engine

Score Viewer

User base

User base

Outlook More users ● Music libraries around the world ● Broadcasting organizations ● Publishers

Technologies ● Google App Engine, Google Storage ● Amazon SQS, EC2, S3 ● Virtualization ● Hadoop, HBase

Workloads ● OMR – optical music recognition ● commercial software – in VMs ● own data-mining algorithms – in Hadoop ● Data mining ● Evolution of music ● Search log data ● Other CPU-intensive IR tasks ● syncrhonizing audio and video with scores