How can musical information be retrieved from a musical document?
There are two ways of accessing the musical content of a work.
1. From an audio input
Using various audio-based data extraction tools such as Aubio (http://aubio.org/), Yaafe (https://github.com/Yaafe/Yaafe) and Vamp plugins (www.vamp-plugins.org).
Diagram of Vamp Plugin (www.vamp-plugins.org) showing data extraction from audio inputs
Once the data has been extracted, a whole world of brand new analytical possibilities opens up for musicologists, ethnomusicologists and researchers interested in the phenomenon of music. They can analyse musical data to find relationships in the musical structure, such as the most frequent chord resolutions in a particular style or how a series is developed in a twelve-tone work, or look at the differences between hundreds of performances of the same piece of music. One of the commonest techniques in this kind of analysis is to convert the data obtained into visualisations, such as those provided, for example, by the Sonic Visualiser Free Software, whose possibilities go beyond conventional sound recordings: www.sonicvisualiser.org
The Digital Music Lab project (http://dml.city.ac.uk/) contains several examples of musicological research using Big Data, such as its DML Vis interface which is available on-line (http://dml.city.ac.uk/vis/). This application enables users to explore, analyse and compare sound recordings in three major collections (the British Library Sound Archives, CHARM and I like Music) by viewing the duration of movements, the notes which predominate in terms of length, geographical provenance, chords, etc.
Screenshot of the DML Vis Interface2. From an image
New tools have emerged in recent years which allow musical notation to be codified from its digitised image. Optical Music Recognition (OMR) allows large quantities of musical information to be accessed and processed in digital libraries.
Until new optical recognition technologies appear, the key to reducing conversion errors would appear to involve subsequent processing of the codified data.
As a solution to correct the results obtained after performing OMR, the researchers and research groups we mention below propose processing based on the following strategies:
- Using various software applications for OMR and subsequently combining their results in order to improve the final score (Víctor Padilla et. al, Improving OMR for Digital Music Libraries with Multiple Recognisers and Multiple Sources, DLfM 2014, September 12th, 2014, London, UK).
- Segmenting scores into bars and using dynamic programming techniques on the data obtained (Jin Rong and Christopher Raphael, Interpreting rhythm in optical music recognition, 13th International Society for Music Information Retrieval Conference [ISMIR 2012]).
- c) Analysing rhythmical similarities throughout scores (and therefore paying attention to the cultural style of repertoires) and subsequent correction of errors by applying probability models (Maura Church and Michael Scott Cuthbert, Improving rhythmic transcriptions via probability models applied post-OMR, 15th International Society for Music Information Retrieval Conference, 2014).
In addition, with manuscript scores (which we assessed with particular care in the tests) the situation is made even more difficult, if that is possible, by the fact that the musical notation itself is not standardised. This problem is in fact common in the repertoire we used as a case study in our library for this project – vocal music for the stage. For example, if one of the pages does not keep the key and time signatures, the musical writing will be so diverse and so specific that it may produce a conversion which is unusable in practical terms. At present, the conversion represented by a digitised manuscript score is completely ambiguous to the OMR software and requires post-conversion manual correction to obtain an accurate representation of the musical notation.
Trio Cui improviso fulmin from the opera by Zulima de Marcos Portugal,JMPM-015 (Biblioteca Fundación Juan March)