We conduct research on signal processing and machine learning
using multimedia information such as audio and images.

A biometric authentication technology that uses voice as biological information to verify whether an input voice belongs to the registered individual. We aim to realize highly accurate and secure speaker verification based on deep learning speaker embeddings.

A technology for determining whether input speech was actually produced by a human. We are developing robust spoofed-speech detection models using deep learning and speech signal processing, including database construction.

Technologies that prevent a speaker from being identified from their speech while keeping the speech useful for downstream applications. We develop attacks that identify the original speaker from anonymized speech, aiming at reliable evaluation methods.