Dataset Open Access

Semantic FM Synth Dataset

Ben Hayes

A dataset of interconnected FM synthesiser patches with comparative semantic tags, as well as rich metadata. This data was collected as part of a perceptual study on the semantic associations of FM synthesis [1]

Participants created FM synthesiser patches using a three operator architecture in response to semantic prompts. They also rated the magnitude of the difference between the created patch and the starting patch in terms of 27 semantic descriptors.

Dataset contents

Audio files are presented in two folders: audio_long contains 4 second renderings of the synth patches with a 3-second gate followed by 1-second of release. audio_short contains 1.25 second renderings of the synth patches with a 1-second gate followed by 0.25-seconds of release. Audio is all presented in 16-bit 44.1kHz PCM WAV files.

File names correspond to the synth_id field from metadata.all.csv and metadata.analysis.csv.

audio_long/
  - 00c4b7cf43ca323183879f32e0d500d0.wav
  - 018bcee372474b8758dab4811651c055.wav
  - ...
audio_short/
  - 00c4b7cf43ca323183879f32e0d500d0.wav
  - 018bcee372474b8758dab4811651c055.wav
  - ...
metadata.all.csv
metadata.analysis.csv
README.md

metadata.all.csv provides metadata for _all_ audio files collected during the study, as well as the 9 reference synth patches used to "seed" the procedure. metadata.analysis.csv provides metadata for only the synthesiser patches analysed in the study, and is provided for the purposes of reproducibility. Participants were included in the analysis only if they met language and age criteria.

Metadata Format

The following metadata is provided (presented in the order of headings in the provided CSV files):

  • synth_id — Unique identifier of synthesiser patch
  • reference_synth_id — Unique identifier of reference synthesiser patch (from which participants started the synthesis process)
  • participant_id — Unique identifier of participant
  • prompt — The semantic prompt given to the participant (e.g. tweak these parameters to make this sound rougher)
  • prompt_descriptor — The root word of the semantic prompt given to the participant
  • prompt_direction — The "direction" (positive or negative) of the prompt
  • delta_* — The amount by which the relevant synthesiser parameter was changed from the starting reference position
  • param_* — The final value of the relevant synthesiser parameter
  • ref_* — The starting reference value of the relevant synthesiser parameter
  • bright to plucky — A comparative semantic rating (on a scale of -10.0 to 10.0) between the starting reference patch and the final created patch
  • note — The MIDI note at which both the reference and created patch were played
  • semantic_factor_* — The patch's score along each of the 5 semantic factors (see [1] for more info)
  • msi_* — The participant's response to Goldsmiths Musical Sophistication Index subscales (see [2] for more info)
  • age — The participant's age
  • country_childhood — The country in which the participant spent the formative years of their childhood
  • country_residence — The country in which the participant resided at the time of participation
  • stft_mag_centroid_median to zero_crossing_rate — A large number of acoustic features extracted on each sound
  • acoustic_component_* — The patch's score along each of 4 principal components extracted from the acoustic features
  • delta_acoustic_component_* — The difference between this patch's score and the reference patch's score on each of the principal components

References

[1] B. Hayes and C. Saitis, ‘There’s more to timbre than musical instruments: semantic dimensions of FM sounds’, presented at the Timbre, Thessaloniki, Greece (Online), 2020, Advance online publication.

[2] D. Müllensiefen, B. Gingras, J. Musil, and L. Stewart, ‘The Musicality of Non-Musicians: An Index for Assessing Musical Sophistication in the General Population’, PLOS ONE, vol. 9, no. 2, p. e89642, Feb. 2014, doi: 10.1371/journal.pone.0089642.

Files (136.2 MB)
Name Size
audio_long.zip
md5:48ba40585589e0f1c32894d8571ee8fc
98.9 MB Download
audio_short.zip
md5:70df62d436b71766c22df530ffd15887
36.2 MB Download
metadata.all.csv
md5:3ef4c582353268514a0e311869b8d5c5
598.8 kB Download
metadata.analysis.csv
md5:e29cef321065e7cfa3a2f3a162c5c668
436.8 kB Download
README.md
md5:368bbe61548fc302ea787c89dbf4485d
4.1 kB Download
93
37
views
downloads
All versions This version
Views 9358
Downloads 3730
Data volume 954.0 MB817.1 MB
Unique views 7552
Unique downloads 2319

Share

Cite as