tomato v0.13.0

tomato

Turkish-Ottoman Makam (M)usic Analysis TOolbox

Introduction

tomato is a comprehensive and easy-to-use toolbox in Python for the analysis of audio recordings and music scores of Turkish-Ottoman makam music. The toolbox includes the state of art methodologies applied to this music tradition. The analysis tasks include:

• Symbolic Analysis: score metadata extraction, score section extraction, score phrase segmentation, semiotic section, and phrase analysis
• Audio Analysis: audio metadata crawling, predominant melody extraction, tonic and transposition identification, makam recognition, pitch distribution computation, tuning analysis, melodic progression analysis
• Joint Analysis: score-informed tonic identification and tempo estimation, section linking, note-level audio-score alignment, predominant melody octave correction, note modeling

The toolbox aims to facilitate the analysis of large-scale audio recording and music score collections of Turkish-Ottoman makam music, using the state of the art methodologies specifically designed for the culture-specific characteristics of this tradition. The analysis results can then be further used for several tasks such as automatic content description, music discovery/recommendation, and musicological analysis.

If you are using tomato in your work, please cite the dissertation:

Şentürk, S. (2016). Computational Analysis of Audio Recordings and Music Scores for the Description and Discovery of Ottoman-Turkish Makam Music. Ph.D. thesis, Universitat Pompeu Fabra, Barcelona, Spain.

For the sake of reproducibility, please also state the version you used as indexed at Zenodo.

For the descriptions of the methodologies in the toolbox, please refer to the papers listed in the References.

tomato v0.13.0

• Refactored the code (in particular metadata and score processing), which was previously ported from different makam analysis libraries in v0.11.0: Pull requests #84 and #94

Tomato in a Nutshell

# import ...
from tomato.joint.completeanalyzer import CompleteAnalyzer
from matplotlib import pyplot as plt

# score input
symbtr_name = 'makam--form--usul--name--composer'
txt_score_filename = 'path/to/txt_score'
mu2_score_filename = 'path/to/mu2_score'

# audio input
audio_filename = 'path/to/audio'
audio_mbid = '11111111-1111-1111-1111-11111111111'  # MusicBrainz Recording Identifier

# instantiate analyzer object
completeAnalyzer = CompleteAnalyzer()

# Apply the complete analysis. The resulting tuple will have
# (summarized_features, score_features, audio_features,
# score_informed_audio_features, joint_features) in order
results = completeAnalyzer.analyze(
    symbtr_name=symbtr_name, symbtr_txt_filename=txt_score_filepath,
    symbtr_mu2_filename=mu2_score_filepath, audio_filename=audio_filepath,
    audio_metadata=audio_mbid)

# plot the summarized features
fig, ax = completeAnalyzer.plot(results[0])
ax[0].set_ylim([50, 500])
plt.show()

You can refer to the jupyter notebooks in demos folder for detailed, interactive examples.

Installation

There are four steps in the installation:

1. Installing tomato
2. Installing Essentia
3. Installing MATLAB Runtime
4. Installing LilyPond (optional)

Installing tomato

tomato requires several packages to be installed. In Linux, you have to install the python3 (or python 2.7, depending on your Python choice), libxml2, libxslt1, freetype and png development packages. The package names might vary in different Linux distributions. In Ubuntu 16.04, you can install these packages by:

sudo apt-get install python-dev libxml2-dev libxslt1-dev libfreetype6-dev libpng12-dev

It is recommended to install tomato and its dependencies into a virtualenv. In the terminal, do the following:

virtualenv env
source env/bin/activate

Then change the current directory to the repository folder and install by:

cd path/to/tomato
python setup.py install

The requirements are installed during the setup. If that step does not work for some reason, you can install the requirements by calling:

pip install -r requirements.txt

If you want to edit files in the package and want the changes reflected, you should call:

cd path/to/tomato
pip install -e .

Installing Essentia

tomato uses several modules in Essentia. Follow the instructions to install the library.

If you are using Python 2.7, then you should link the python bindings of Essentia in the virtual environment:

ln -s path_to_essentia_bindings path_to_env/lib/python2.7/site-packages

Don't forget to change the path_to_essentia_bindings and path_to_env with the actual path of the installed Essentia Python bindings and the path of your virtualenv, respectively. Depending on the Essentia version, the default installation path of the Essentia bindings is either /usr/local/lib/python2.7/dist-packages/essentia or /usr/local/lib/python2.7/site-packages/essentia.

Installing MATLAB Runtime

The score phrase segmentation, score-informed joint tonic identification and tempo estimation, section linking and note-level audio-score alignment algorithms are implemented in MATLAB and compiled as binaries. They need MATLAB Runtime for R2015a (8.5) to run. You should download and install this specific version (links for Linux and Mac OSX).

We recommend you to install MATLAB Runtime in the default installation path, as tomato searches them automatically. Otherwise, you have to specify your own path in the MATLAB Runtime configuration file, tomato/config/mcr_path.cfg.

Installing LilyPond

If you want to convert the music scores to SVG format, LilyPond is a good choice, because it adds a mapping between each musical element in the LilyPond file and in the related SVG.

To install LilyPond in Mac OSX, simply go to the Download page in the LilyPond website and follow the instructions for your operating system.

In most Linux distributions, you can install LilyPond from the software repository of your distribution. However, the version might be outdated. If the version is below 2.18.2, we recommend you to download the latest stable version from the LilyPond website. If you had to install LilyPond this way, you should enter the LilyPond binary path to the "custom" field in tomato/config/lilypond.cfg (the default location is $HOME/bin/lilypond).

Documentation

Coming soon...

License

The source code hosted in this repository is licensed under Affero GPL version 3.

Any data (the music scores, extracted features, training models, figures, outputs etc.) are licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

FAQ

1. The notes aligned by JointAnalyzer.align_audio_score(...) seems shifted. What is the problem?

   Your audio input is probably a compressed format such as mp3. There are typically shifts between different decoders (and even different versions of the same decoder) when they decode the same compressed audio file. In the predominant melody extraction step (AudioAnalyzer.extract_pitch(...)), Essentia has to decode the recording for processing. You observe a shift when the application you use another decoder.

   These shifts are typically small (e.g. 50 samples ~1ms), so they are not very problematic. Nevertheless, there is no guarantee that the shift will be bigger. If you need "perfect" synchronization, you should use an uncompressed format such as wav as the audio input.

   Note: In demos, we use mp3, because it would be too bulky to host a wav file.

2. Which operating systems are supported?

   The algorithms, which are written purely in Python, are platform-independent. However, compiling Essentia in Windows is not straightforward yet. Therefore we have only compiled the MATLAB binaries for Mac OSX and Linux.
   If you have compiled Essentia for Windows somehow or if you have any OS-specific problems, please let us know by submitting an issue.

3. What are the supported Python versions?

   tomato supports both Python 2.7 and Python 3.

4. Where are the MATLAB binaries?

   The binaries are not stored in tomato because they are relatively big. It would take too much space to store them here, including the versions introduced in each modification. Instead, the binaries are provided within the releases of the relevant packages. The binaries are downloaded to tomato/bin during the installation process of tomato.
   Please refer to tomato/config/bin.cfg for the relevant releases.

5. ScoreConverter says that "The lilypond path is not found". How can I fix the error?

   There can be similar problems regarding this issue:

   • The user-provided file path (the music score input) does not exist.

     Check your input MusicXML-score path.

   • LilyPond is not installed.

     Install the latest stable version for your OS.

   • The binary path exists but it is not used.

     The path is not searched by the defaults defined in tomato/config/lilypond.cfg. Add the path of the LilyPond binary to the configuration file.

6. Is tomato a fruit or vegetable?

   It has a culture-specific answer.

Authors

Sertan Şentürk
[email protected]

Acknowledgements

We would like to thank Harold Hagopian, the founder of Traditional Crossroads, for allowing us to use Tanburi Cemil Bey's performance of Uşşak Sazsemaisi in our demos.

References

The toolbox has been realized as part of the thesis:

[1] Şentürk, S. (2016). Computational Analysis of Audio Recordings and Music Scores for the Description and Discovery of Ottoman-Turkish Makam Music. Ph.D. thesis, Universitat Pompeu Fabra, Barcelona, Spain.

The methods used in the toolbox are described in the papers:

Score Phrase Segmentation
[1] Bozkurt, B., Karaosmanoğlu, M. K., Karaçalı, B., Ünal, E. Usul and Makam driven automatic melodic segmentation for Turkish music. Journal of New Music Research. Vol. 43, Iss. 4, 2014

Score Section Extraction; Semiotic Section and Phrase Analysis
[2] Şentürk S., and Serra X. (2016). A method for structural analysis of Ottoman-Turkish makam music scores. In Proceedings of 6th International Workshop on Folk Music Analysis (FMA 2016), pages 39-46, Dublin, Ireland

Audio Predominant Melody Extraction
[3] Atlı, H. S., Uyar, B., Şentürk, S., Bozkurt, B., and Serra, X. (2014). Audio feature extraction for exploring Turkish makam music. In Proceedings of 3rd International Conference on Audio Technologies for Music and Media (ATMM 2014), pages 142–153, Ankara, Turkey.

Audio Pitch Filter
[4] Bozkurt, B. An Automatic Pitch Analysis Method for Turkish Maqam Music. Journal of New Music Research. 37(1), 1-13.

Audio Tonic and Transposition Identification, Makam Recognition, Pitch Distribution Computation, Tuning Analysis
[6] Gedik, A. C., and Bozkurt, B. (2010). Pitch-frequency histogram-based music information retrieval for Turkish music. Signal Processing, 90(4), 1049-1063.
[7] Bozkurt, B. 2008, An automatic pitch analysis method for Turkish maqam music, Journal of New Music Research 37 1–13.
[8] Chordia, P. and Şentürk, S. (2013). Joint recognition of raag and tonic in North Indian music. Computer Music Journal, 37(3):82–98.

Audio Tonic Identification from the Last Note
[9] Atlı, H. S., Bozkurt, B., and Şentürk, S. (2015). A method for tonic frequency identification of Turkish makam music recordings. In Proceedings of 5th International Workshop on Folk Music Analysis (FMA 2015), pages 119–122, Paris, France.

Audio Melodic Progression (Seyir) Analysis
[10] Bozkurt B., Computational analysis of overall melodic progression for Turkish Makam Music, in Penser l’improvisation edited by Mondher Ayari, pp. 289-298, ISBN: 9782752102485, 2015, Delatour France, Sampzon.

Score-Informed Audio Tonic Identification
[11] Şentürk, S., Gulati, S., and Serra, X. (2013). Score informed tonic identification for makam music of Turkey. In Proceedings of 14th International Society for Music Information Retrieval Conference (ISMIR 2013), pages 175–180, Curitiba, Brazil.

Score-Informed Audio Tempo Estimation
[12] Holzapfel, A., Şimşekli U., Şentürk S., and Cemgil A. T. (2015). Section-level modeling of musical audio for linking performances to scores in Turkish makam music. In Proceedings of 40th IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2015), pages 141–145, Brisbane, Australia.

Audio-Score Section Linking
[13] Şentürk, S., Holzapfel, A., and Serra, X. (2014). Linking scores and audio recordings in makam music of Turkey. Journal of New Music Research, 43:34–52.

Note-Level Audio-Score Aligment
[14] Şentürk, S., Gulati, S., and Serra, X. (2014). Towards alignment of score and audio recordings of Ottoman-Turkish makam music. In Proceedings of 4th International Workshop on Folk Music Analysis (FMA 2014), pages 57–60, Istanbul, Turkey.

Score-Informed Audio Predominant Melody Correction; Note Modeling
[15] Şentürk, S., Koduri G. K., and Serra X. (2016). A score-informed computational description of svaras using a statistical model. In Proceedings of 13th Sound and Music Computing Conference (SMC 2016). pages 427-433, Hamburg, Germany