Timing Data

The basic unit of time in our harmonic analyses and melodic transcriptions is the measure (or bar), which is a relative quantity. We also measured time in absolute terms, by tracking the exact location of every bar line in the original audio file. This timing data could be useful for exploring aspects of meter and rhythm that depend on absolute time, such as the pacing of harmonies or the speed of a vocal delivery, that have yet to receive much examination in the field of music theory. It also allows our harmonic analyses and melodic transcriptions to be used with music information retrieval projects (more on this below).

The timing data files were created using the Sonic Visualizer software developed at the Centre for Digital Music at Queen Mary, University of London. Each audio file was loaded into the software, after which rough measure locations were placed during an initial pass by tapping on the computer keyboard. These rough locations were then fine-tuned by a second pass, using a combination of listening for downbeats and seeing downbeats through the visual representation of the waveform.

The timing data is distributed via tab-delimited files. The first column is the absolute time (in seconds) from the beginning of the original audio file. (Note that often there is a short period of silence, such as a second or so, at the beginning of an audio file before the first actual sound is heard.) The second column is the measure number. The first measure is always considered to be measure 0. In the case of an upbeat or anacrusis event, we notate a full measure (padded with rests prior to the first event) instead of a partial measure as used in traditional notation. (Note that we often ignore brief instrumental pickups, although we are careful to account for all melodic and all significant harmonic anacruses.) In songs with a notated pickup, the timing data file will typically begin with measure 1, since the location of the beginning of measure 0 does not exist within the original audio file. Measure numbers in the timing data align with those output from the expand6.c program.

You can browse the timing data files individually, or download the entire collection as a zip file:

This timing data is incorporated into our "timed chordlist" and "timed notelist" formats, which can be downloaded on our harmonic analyses and melodic transcriptions pages. These timed lists were created using the add-timings.pl program in combination with the expand6.c (for the harmonic analyses) and process-mel5.pl (for the melodic transcriptions) programs.

We also provide a tab-delimited list of the sources of the original audio files used in creating the timing data. This columns in this list are: 1) the filename convention we used, 2) the artist, 3) the name of the album from which the recording was taken, and 4) any relevant publication information (publisher, year) for those albums that may require clarification (e.g., greatest hits compilations). All of the audio files were taken from compact disc (CD) versions of the releases.

Given that the corpus currently contains 200 songs, with very few songs available from the same album, collecting the original audio files was a time-consuming task. The difficulty in compiling this collection of audio files makes using our timing data with music information retrieval projects somewhat prohibitive, since it is doubtful that another researcher could recreate the same collection of audio files without spending a similar amount of time and effort. If you are interested in using our timing data with a music information retrieval project, and/or you have ideas about how to overcome this issue, please do not hesitate to contact us.