A year-long deep-sea soundscape dataset off Minamitorishima Island

This dataset includes acoustic data associated with the deep-sea soundscape off Minamitorishima Island, along with Python codes for using the audio data in a Google Colab notebook.

Recording Location

The audio recordings were collected from the seafloor east of Takuyo-Daigo Seamount (22°59.9' N, 154°24.5' E, depth 5552 m), approximately 150 km south of Minamitorishima Island.

Recording System

The data collection used a 6000 m rated icListen HF hydrophone (model SB60L-ETH, Ocean Sonics, Canada) and a Gordon Smart Recorder (Ocean Sonics, Canada). The hydrophone was mounted atop an Edokko Mark I Type 365 (Okamoto Glass, Japan) and connected to two pressure-resistant glass spheres housing the Gordon Smart Recorder and battery packs. The recording system has a hydrophone sensitivity of -170 dB re 1V μPa-1, with a frequency response from 10 Hz to 200 kHz (±6 dB bandwidth). See details in Onishi et al. (2023).

Configuration of Audio Recording

(1) Duty Cycle: 2-min recording every 4 hours (2 AM, 6 AM, 10 AM, 2 PM, 6 PM, and 10 PM, UTC+0). (2) Sampling Rate: 512 kHz. (3) File Format: WAV. (4) Audio Gain: 0 dB. (5) High Pass Filter: Off.

Field Deployment

The Edokko Mark I was deployed on March 13, 2020, during the R/V KAIREI cruise KR20-E01C and retrieved on April 9, 2021, during the R/V KAIREI cruise KR21-04C. However, effective recordings only covered the period between March 13, 2020 and March 5, 2021.

Data Structure

  • Audio: Folders of audio recordings, with folder names indicating the date of recording (yyyymmdd). Each folder contains 24 wav files recorded in one day, with file names in the format BPW33333_yyyymmdd_HHMMSS. All timestamps on the folders and wav files are in UTC+0 (local time is UTC+10).
  • LTS: A mat file containing the long-term spectrogram of the year-round seafloor recordings. There are two types of LTS: median-based and mean-based. The median-based LTS was obtained by measuring the median of power spectral densities in each frequency bin, while the mean-based LTS was obtained by measuring the mean of power spectral densities in each frequency bin. Both types have a frequency resolution of 10 Hz and a time resolution of 1 minute. Use the soundscape_IR Python package or MATLAB to open the mat file.
  • Transient Sound: A mat file containing the spectral features associated with high-intensity transient sounds, learned using semi-supervised source separation. Use the soundscape_IR Python package or MATLAB to open the mat file.
  • Model: Mat files containing source separation models for fish choruses and ambient noise. Use the soundscape_IR Python package to load these models for performing source separation.


This work was conducted under the following ocean programs supported by the Cross-ministerial Strategic Innovation Promotion Program: Innovative Technologies for Exploration of Deep-Sea Resources from 2018 to 2022 (Lead agency: Japan Agency for Marine-Earth Science and Technology).

Associated Publication

Lin, T.-H., Kawagucci, S. (2023) Acoustic twilight: A year-long seafloor monitoring unveils phenological patterns in the abyssal soundscape. Limnology and Oceanography Letters, early view.



Wikidata 關鍵字

  • Q22712980
  • Q1358257


  • 原始碼
  • 影音資料
  • 科學與統計資料
語言 英文 (eng)


起始時間 2020-03-13
結束時間 2021-03-05
空間範圍.X.min 154.408333
空間範圍.X.max 154.408333
空間範圍.Y.min 22.998333
空間範圍.Y.max 22.998333


產製者 Tzu-Hao Lin, Shinsuke Kawagucci
聯絡人 Tzu-Hao Lin
聯絡人的電子郵件 schonkopf@gmail.com