[AUDIO] MTG - Query-by-Humming (QBH)
Check the original page for updates here
Audio Files Included
Collection_Full.csv
This file contains meta-data information about the full collection of songs against which the queries were evaluated in the experiments described in the aforementioned article. The collection was compiled in the context of:
- Song ID: a unique and arbitrary identifier given to each recording. NOTE: the ID is unique for every recording, meaning different versions (recordings) of the same song will have different IDs.
- Title: the title of the song. Note that recordings of the same song may have different titles, for example if they are sung in different languages.
- Artist: the performing artist for this specific recording.
- Original Artist: the composer of the song.
- Canonical: whether this recording is the canonical version of this song. If it is, the field is set to 'YES'. Otherwise (i.e. not the canonical version) the field is left empty. NOTE: this labeling was done by the authors of this dataset and is completely subjective.
- Class label: a unique label for all recordings of the same musical piece. This is the only field guaranteed to be identical for all recordings which are considered by the authors of the dataset to be versions of the same musical piece.
Collection_Canonicals.csv
This file contains meta-data information about the collection of canonical songs against which the queries were evaluated in the experiments described in the aforementioned article. Unlike the full collection, this collection includes only one version of every musical piece, the one considered by the authors of the dataset to be the canonical version of the piece. This means this collection is a SUBSET of the full collection above. NOTE: the selection of canonical songs was done by the authors of the dataset and is completely subjective. In total 481 recordings are described, and for each recording there are six columns, as for the full collection. Since this collection is a subset of the full collection, the column values for a recording with certain Song ID in Collection_Canonicals.csv will be identical to the column values for the same recording (i.e. same Song ID) in Collection_Full.csv.
Queries.csv
- Filename: the name of the audio file.
- Query ID: the ID of the query.
- Song ID: the ID of the song to which the query corresponds. NOTE: use this ID to match the query to the song to which it corresponds in Collection_Full.csv and Collection_Canonicals.csv.
- Original Artist: the composer of the song to which the query corresponds.
- Class label: a unique label for all recordings of the same musical piece. This is the only field guaranteed to be identical for all recordings (whether query or song) which are considered by the authors of the dataset to be versions of the same musical piece.
Download
Please go to the Download page.