Chris@157: 
Chris@157: 
Chris@157: Silvet: Shift-Invariant Latent Variable Transcription
Chris@157: =====================================================
Chris@157: 
Chris@157: A polyphonic music transcription plugin.
Chris@157: 
Chris@157:     http://code.soundsoftware.ac.uk/projects/silvet
Chris@157: 
Chris@157: Silvet is a Vamp plugin (http://vamp-plugins.org) for automatic music
Chris@157: transcription, using the method of "A Shift-Invariant Latent Variable
Chris@157: Model for Automatic Music Transcription" by Emmanouil Benetos and
Chris@157: Simon Dixon (see CITATION file).
Chris@157: 
Chris@157: 
Chris@157: What does it do?
Chris@157: ----------------
Chris@157: 
Chris@157: Silvet listens to audio recordings of music and tries to work out what
Chris@157: notes are being played.
Chris@157: 
Chris@157: To use Silvet, you need a Vamp plugin host such as Sonic Visualiser
Chris@157: (http://sonicvisualiser.org). How to use the plugin will depend on the
Chris@157: host, but in the case of Sonic Visualiser, you should load an audio
Chris@157: file and then run Silvet Note Transcription from the Transform
Chris@157: menu. This will add a note layer to your session with the
Chris@157: transcription in it, which you can play back or export as a MIDI file.
Chris@157: 
Chris@157: 
Chris@157: How good is it?
Chris@157: ---------------
Chris@157: 
Chris@157: Reasonable for recordings that suit it: chamber music, solo piano,
Chris@157: acoustic jazz, etc. But the range of music that works well is quite
Chris@157: limited at this stage.
Chris@157: 
Chris@157: Silvet uses a probablistic latent-variable estimation method to
Chris@157: decompose a Constant-Q time-frequency matrix into note activations
Chris@157: using a set of spectral templates learned from recordings of solo
Chris@157: instruments. This means its performance is dominated by the
Chris@157: correspondence between its instrument templates and the sounds present
Chris@157: in the recording.
Chris@157: 
Chris@157: The method performs quite well (70-85% of notes identified correctly)
Chris@157: for clear recordings that contain only instruments with a good
Chris@157: correspondence to the known templates. In these cases its performance
Chris@157: becomes limited by the note decomposition step, clustering pitch
Chris@157: probabilities into note events, which is still fairly simplistic.
Chris@157: 
Chris@157: Silvet does not yet contain any vocal templates, or templates for
Chris@157: typical rock or electronic instruments. So it will usually perform
Chris@157: very poorly with pop and rock music, although the results can be
Chris@157: interesting anyway. Silvet also makes no attempt to transcribe
Chris@157: percussion.
Chris@157: 
Chris@157: For a formal evaluation, please refer to the 2012 edition of MIREX,
Chris@157: the Music Information Retrieval Evaluation Exchange, where the basic
Chris@157: method implemented in Silvet formed the BD1, BD2 and BD3 submissions
Chris@157: in the Multiple F0 Tracking task:
Chris@157: 
Chris@157: http://www.music-ir.org/mirex/wiki/2012:Multiple_Fundamental_Frequency_Estimation_%26_Tracking_Results
Chris@157: 
Chris@157: 
Chris@157: Authors and copyright
Chris@157: ---------------------
Chris@157: 
Chris@157: The method implemented in Silvet is by Emmanouil Benetos, see
Chris@157: "A Shift-Invariant Latent Variable Model for Automatic Music
Chris@157: Transcription" by Emmanouil Benetos and Simon Dixon (CMJ 2012).  If
Chris@157: you make use of this software for academic purposes, please cite this
Chris@157: publication (see the the CITATION file for BibTeX).
Chris@157: 
Chris@157: The plugin code is by Chris Cannam and Emmanouil Benetos and is
Chris@157: Copyright 2014 Queen Mary, University of London. It is distributed
Chris@157: under the GNU General Public License: see the file COPYING for
Chris@157: details.
Chris@157: 
Chris@157: If you make use of this software for any public or commercial purpose,
Chris@157: we ask you to kindly mention the authors and Queen Mary, University of
Chris@157: London in your user-visible documentation. We're very happy to see
Chris@157: this sort of use but would much appreciate being credited, independent
Chris@157: of the requirements of the software license itself.
Chris@157: