HOTGAME-MFCC: A corpus of early HOuse and Techno music from Germany and AMErica — Mel-Frequency Cepstral Coefficients

HOTGAME-MFCC, is a corpus of Mel Frequency Capstral Coefficients (MFCCs) 1-13, as well as spectral centroid, spectral rolloff, spectral flatness, spectral bandwidth, Zero Crossing Rate (ZCR) and tempo in terms of Beats Per Minute (BPM) extracted from HOuse and Techno music from Germany and AMErica until 1994. In the corpus, 4,521 German tracks from 97 dance music labels and 4,256 US-American tracks from 85 record labels are included.The corpus is almost identical to the HOTGAME-Corpus. The music was stored as (lossly) compressed stereo audio files with a sample rate of 44,100 Hz. Inclusion in the HOTGAME corpus followed several rules: Only Music between 1984 and 1994 is included Ancestors of House and Techno – like (Italy) Disco, Industrial, and Electropop – are not included Music genres that evolved in parallel with House and Techno – such as Electronic Body Music (EBM), High-NRG, Dark Wave, New Wave, Neue Deutsche Welle (NDW), Synth-Pop – are not included Music with stronger influences from other genres – such as Electro Funk, Freestyle, Miami Bass, and Indie-Electro – are not included Descendants of House and Techno – such as Hip House, Eurodance, Trance, Acid, Hardcore and Gabber – are included Artists are assigned to the nation in which the producer was raised HOTGAME consists of CSV files that contain 18 low level features. The features were extracted from non-overlapping time frames of 2,048 samples, i.e., every 50 ms. The only exception are the BPM, which are extracted once for each track. The files are named according to the following convention: artist - title (version) (year).format.csv The year refers to the release year of the record (in rare cases re-releases). They are stored in the sub-folders germany and usa, which can be found in techno-mfccbpm.zip. The source code for the feature extraction is provided on Github in the repository link. The purpose of HOTGAME-MFCCis to provide more common low-level audio features than the recording studio features in the original HOTGAME dataset. The purpose is to enable researchers to study the development of early house and techno music in the United States of America and Germany from an objective, audio-based viewpoint, i.e., the music itself. With over 9,000 tracks, big data methods, such as inferential statistical analysis and machine learning, are appropriate. Such analyses can validate statements, narratives, and conclusions made by time witnesses that can be found in documentaries, biographies, and qualitative studies of electronic dance music, electronica, etc. HOTGAME-MFCC does not contain any audio material. Citation When you use the data set, please cite the respective paper: Ziemer, Tim, HOTGAME: A Corpus of Early House and Techno Music from Germany and America, Metrics 2025, 2(2): 8, https://doi.org/10.3390/metrics2020008 One analysis of the HOTGAME-MFCC corpus is: Ziemer, Tim, Mel-Frequency Cepstral Coefficients and Recording Studio Features for the Analysis of Producer-Driven Music, preprints.org 2026, https://doi.org/10.20944/preprints202603.0656.v1 One analysis of the original HOTGAME corpus is: Ziemer, Tim and Linke, Simon, An Audio-Perspective on the Divergent Paths of Techno in Germany and the United States, Transactions of the International Society for Music Information Retrieval 9(1), https://doi.org/10.5334/tismir.324, pp. 264-279 Be aware of the limitations of the provided low-level features: While they provide sufficient and partly orthogonal dimensions for supervised machine learning algorithms to fit to any desired outcome, they cannot be interpreted well, as they neither relate to the way the music was made, nor to the way it is perceived (except for the BPM which relates to both).