# Activity Recognition from Daily-Life Sounds Using Unsupervised Learning with Dirichlet Multinomial Mixture Models

## TL;DR

This paper presents an unsupervised method for recognizing daily activities from household sounds to support elderly care.

## Contribution

The novelty lies in using Dirichlet multinomial mixture models for activity recognition with minimal labeled data.

## Key findings

- The model effectively clusters activities from audio codec codes without supervision.
- Handling multiple sound directions improves the accuracy of activity recognition.
- The approach reduces the need for labeled data and user input.

## Abstract

To support ambient assisted living for the elderly living alone, we investigate a method for recognizing daily activities from household sounds. To reduce the cost of building an activity-recognition model, we adopt an unsupervised learning approach based on a Dirichlet multinomial mixture model. The model represents the generative process of neural audio codec codes conditioned on latent activities. We further extend the model to handle multiple streams of codes corresponding to different sound directions. This extension enables the formation of more accurate activity clusters, partly because code occurrence patterns exhibit burstiness. The proposed approach is expected to serve as a key component for constructing an activity recognition system that requires minimal labeled data and a small number of user inquiries.