SocialReverb Data

This page hosts all the data collected by SocialReverb, a project by Northwestern University Interactive Audio Lab.

SocialReverb is a project to crowdsource a vocabulary of audio descriptors for reverberation. We can use the data collected to map these words onto actions that can be taken by a parametric reverberator. This data can be used to make new sorts of interactions possible, such as simply asking a tool to "make the sound boomy".

To date we have collected 3388 unique words covering 256 reverberation settings, collected from 658 participants over 1397 sessions. We provide the data as csv files, as well as the associated audio with SocialReverb.

The Audio

  1. Get the piano sample here.
  2. Get the guitar sample here.
  3. Get the drums sample here.

The Data

Get each individual contribution to SocialReverb here.

Below is an explanation of each header in "contributions.csv".

  1. likert: the answer to the prompt "Indicate how much the reverb affects the audio". 0 - not at all, 1 - somewhat, 2 - moderately, 3 - strongly, 4 - very strongly.
  2. effectTime: time participant spent listening to the audio with reverberation on.
  3. didnotagree: words that the participant disagreed described the audio.
  4. rawwords: words participant contributed freely in first part of survey. This the raw input to the form, before any spell check is run.
  5. agreed: words that the participant agreed described the audio.
  6. when: time and date of contribution.
  7. userid: unique identifier for a single participant.
  8. effect: what effect was applied. In this case, it will always be "reverb".
  9. param: what parameters was the associated reverberation set at? These are the actual controls on the reverberator, not the measures (explained in next section).
  10. quiet: answer to the question "Is the listening environment quiet?" (yes, no).
  11. sessionid: unique identifier for a single session of SocialReverb.
  12. ind: index of reverberation setting used. Use this to find more information about the reverberation setting in the next file "reverbs.csv".
  13. cleanTime: time participant spent listening to the audio without any effects applied to it.
  14. words: words participant contributed freely after spell checking.
  15. audioFile: what audio file was the participant listening to? (/static/audio/{piano, guitar, drums}.mp3)
  16. listen: what kind of listening environment was the participant in? (headphones, standalone, laptop/tablet/phone, other).

Get information about each reverberation setting in SocialReverb here.

Below is an explanation of each header in "reverbs.csv".

  1. count: how many contributions are associated with this reverberation setting?
  2. index: what is the index of this reverberation? Associated with "ind" entry in "contributions.csv".
  3. impulsemeasures: the measures of the impulse response of the reverberation setting (reverberation time in seconds, echo density, clarity in decibels, central time in seconds, frequency cutoff in Hz).
  4. parameters: the parameters to set the reverberator to make this reverberation setting happen (delay of comb filters in seconds, gain of comb filters between 0 and 1, delay between channels in seconds, frequency cutoff in Hz, gain of reverberation as any positive number).

The reverberator developed in SocialReverb is created using the Web Audio API. Details of its construction can be found in this paper.