Main Researcher(s): Alejandra Gomez Ortega, Wiebke Toussaint
Contact Email: firstname.lastname@example.org
Affiliation: PhD Candidate, TU Delft
Duration: April 25, 2022 to July 15, 2022
Voice biometrics are used in voice assistants, smart speakers, and call centers to identify people from their voices. Like automated speech processing and face recognition, voice biometrics rely on advanced AI and deep learning techniques. To evaluate bias in and the reliability of voice biometrics, it is necessary to gather evaluation datasets that represent real-life usage contexts. However, these datasets are not readily available. We encourage voice assistant users to donate their voice interaction data to support research into the bias of voice-driven systems.
Which data is required?
Speech Records, Speaker Metadata.
How will the data be used?
We will construct a speaker verification evaluation dataset that reflects real-life usage conditions of voice biometrics technology in voice assistants.
How to donate?
Go to takeout.google.com and log in with your Google credentials if necessary. Under ‘select data to include’ click on ‘deselect all’. Scroll down until you find ‘my activity’ and select it. Click on ‘all activity data included’, this should open a pop-up where you can choose specific activity data to export. Click on ‘deselect all’ and select ‘assistant’. Then click on ‘multiple formats’, this should open a pop-up where you can choose the format for your download. Under ‘activity records’ click on JSON and click OK. Scroll down and click on the ‘next step' button. Finally, under ‘choose file type, frequency & destination’ select ‘export once’, ‘.zip’, and ‘2GB’ and click on the ‘create export’ button. You will receive an email with your data, download the .zip file from your email and upload it below.
Detailed instructions can be found here
Step 1: Upload Your Data
This project is closed. At this time you cannot upload your data. We invite you to explore other projects.