Bando Lab

Bando Lab is a sub-team of the Social Intelligence Research Team, Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology (AIST), and is also affiliated with the Intelligent Acoustics Laboratory at the University of Tsukuba through the Cooperative Graduate School Program.

We focus on intelligent systems centered on machine listening, including sound event detection (SED) and distant automatic speech recognition (DASR). Our current research emphasizes self-supervised learning and large-scale pretraining to establish frameworks robust in complex real-world conditions.

Contact #

2-4-7 Aomi, Koto-ku, Tokyo 135-0064, Japan
AIST Waterfront Annex 8F

y.bando [at] aist.go.jp

Recent

M3L: Multi-Modal Machine Listening Toolkit ↗ ↖

M3L is an open-source toolkit for multi-modal machine listening, providing infrastructure and recipes to build, train, and evaluate models that combine audio with other modalities.

SBSS: Scalable Blind Source Separation Toolkit ↗ ↖

SBSS is a Python-based, research-oriented toolkit for scalable blind source separation (BSS) that includes end-to-end components such as neural FCA and neural FastFCA.

Aiaccel ↗ ↖

Aiaccel is an open-source Python toolkit designed to accelerate machine learning research, especially on high-performance computing (HPC) clusters such as ABCI.

Onset-and-Offset-Aware Sound Event Detection via Differentiable Frame-to-Event Mapping

2 mins

SED-HSMM couples a neural acoustic model with a hidden semi-Markov model to deliver event-wise sound detection with precise boundaries.

LEAD Dataset ↗ ↖

The LEAD dataset provides strong labels for sound events, in which each clip has 20 different annotations. It allows us to investigate how annotations vary among different annotators and develop SED models that are robust to the variations.