International Audio Laboratories Erlangen

SMD MIDI-Audio Piano Music

Saarland Music Data provides audio recordings along with perfectly synchronized MIDI files for various piano pieces. The pieces were performed by students of the Hochschule für Musik Saar on a hybrid acoustic/digital piano Yamaha Disklavier. The Disklavier allows for capturing key and pedal movements of the piano while playing. This information, which can be stored in a MIDI file, yields an accurate annotation of the corresponding audio recording in form of a symbolic description of all played musical note events. The SMD MIDI-Audio pairs constitute a valuable dataset for various music analysis tasks such as music transcription, performance analysis, music synchronization, audio alignment, or source separation.

Copyright

If you publish results obtained using this dataset, please cite:

Meinard Müller, Verena Konz, Wolfgang Bogler, Vlora Arifi-Müller: Saarland Music Data (SMD). In Late-Breaking and Demo Session of the 12th International Conference on Music Information Retrieval (ISMIR), 2011. [pdf] [bib]

Saarland Music Data (SMD) by Saarland Music Data (SMD) is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.

Naming Convention

For the audio and MIDI files, we use the following naming convention:

Composer_Work_Performer_Version.mp3 Composer_Work_Performer_Version.mid

For the variables, the following conventions are taken into account:

• Composer: Last name of composer such as Beethoven or Mozart.
• Work: Op (or similar) + 3 digits (for movements: + '-' + 2 digits) such as Op057-02, Op002No2-03, BWV123, KV012 If no work number is available, a suitable acronym is used such as CarminaBurana-05.
• Performer: The performer (orchestra, conductor, soloist) is left anonymous. Instead we use three digit identifier.
• Version: Recording date in the form yyyymmdd or possibly other information ending with -SMD.

Example: Beethoven_Op057-02_013_20090221-SMD.mp3 MP3 recording of Beethoven's piano sonata Opus 57, second movement, played by pianist 013 recorded on Feb. 21, 2009.

Technical Information

All performances were recorded in the studios of the Hochschule für Musik Saar, played by students of piano classes of different levels, on a Yamaha Disklavier model DCFIIISM4PRO. Using two cardioid-condenser microphones fixed over the resonating body of the piano, all performances were directly recorded into Steinberg Cubase 4. Except for trimming the beginnings and ends of the recordings, no further post-processing (filters, effects) was applied to the musical material. From each Cubase project, an audio file (44.1 kHz, stereo) as well as a synchronized standard MIDI file (SMF) were exported. Besides these files, we also provide the audio files as WAV (22.05 kHz, mono) and the MIDI files encoded as CSV files and as WAV files (22.05 kHz, mono) rendered using the Software synthesizer FluidSynth.

Note: The audio and MIDI files are temporally synchronized. In the original data collection (Version 0), we provided MP3 files rather than WAV files. Additionally, there were global offsets between the MIDI and audio versions. On this website, we offer WAV files and corrected MIDI files to compensate for these global offsets. However, note that the overall temporal accuracy is constrained by internal delays of the Disklavier and the speed of sound.

Database

SMD MIDI-Audio Piano Music (Version 1) contains the following data:

• WAV-44: Audio file (44.1 kHz, stereo)
• WAV-22: Audio file (22.05 kHz, mono)
• MIDI: MIDI file
• CSV: Export of MIDI file into CSV format
• WAV-MIDI: MIDI file rendered as audio file (22.05 kHz, mono)

The entire collection can be downloaded as a single ZIP file from ZENODO.

Acknowledgment

We would like to express our gratitude to my former students for their help and feedback. In particular, we thank Johannes Zeitler for his technical support and assistance with data adjustment.