Tools created by A/Vian Eco for the Remote Detection Community!

Sound Optimization for Fading Talkative Signals and Preserving Other Key Environment Noises

Created in collaboration with our amazing programming partner: Jarrett Lubky.

A tool for human privacy protection for open-source environmental acoustic datasets.

SOFTSPOKEN is an open-source, free-to-use tool designed to help researchers find and silence human voices in ecological acoustic datasets. By leveraging a machine learned model, SOFTSPOKEN detects probable human speech sounds while ignoring critical environmental and animal sounds. Users are encouraged review and choose to either silence these detections or export observations as a dataset. The tool is meant to support:

  • sharing audio recordings while ethically preserving human privacy.
  • excluding recordings from analysis that could be affected by human presence.
  • comparing human activity to animal acoustic behaviour.
  • more… tell us about your project!

Key Features:

  • Automated Speech Detection – Uses a trained model to identify human voices while ignoring non-human environmental sounds.
  • No-Code Interface – Intuitive graphical user interface (GUI) allows easy review and selective removal of detected speech segments.
  • Review and Assess – The user interface allows you to review probable human detections by the model. The user can then assess whether to keep or silence the model’s detections.
  • Minimal Data Loss -Ensures that only necessary time intervals with human speech are silenced, preserving the integrity of ecological recordings for open-sourced sharing.
  • Open-Source & Free – Free desktop app available for researchers and practitioners to foster collaboration and innovation in ecoacoustics. If you appreciate python code like we do, our programmer has also generously shared his work under a permissive MIT license.

With SOFTSPOKEN, researchers can confidently contribute to open datasets without compromising privacy, ensuring ethical and responsible data sharing.

Download the SOFTSPOKEN desktop app and see additional information here:

https://github.com/AVianEco/Softspoken

 

github.com/avianeco

We spend a lot of time over here managing large audio/visual datasets for environmental research and consultation. We also teach strategies for assessing the quality of remotely captured sensor data. Check for upcoming in-person courses under News.

Our GitHub is a place where we store R-scripts you might find handy for:

  • Managing and tracking large audio datasets. [See Repo].
  • Manipulating raw outputs from audio analysis programs to optimize data available in secondary analysis (e.g., R, excel, etc.) [See Repo].
  • Softspoken [See Repo].

All of these scripts are permissively licensed for you to download, use, modify to your preference, even for commercial purposes. The data manipulation R-scripts contain explanatory comments to help walk through their use and outputs.

Stay tuned, more useful scripts to come in future updates.