Age Group Classification with Speech and Metadata Multimodality Fusion

TL;DR

This paper introduces a multimodal approach combining speech and user metadata to improve age group classification, achieving state-of-the-art accuracy in identifying children from short audio commands.

Contribution

It presents a novel ensemble method that fuses speech and metadata data, significantly enhancing child detection accuracy over previous techniques.

Findings

9.2% absolute improvement over baseline

State-of-the-art performance achieved

Effective multimodal fusion for age classification

Abstract

Children comprise a significant proportion of TV viewers and it is worthwhile to customize the experience for them. However, identifying who is a child in the audience can be a challenging task. Identifying gender and age from audio commands is a well-studied problem but is still very challenging to get good accuracy when the utterances are typically only a couple of seconds long. We present initial studies of a novel method which combines utterances with user metadata. In particular, we develop an ensemble of different machine learning techniques on different subsets of data to improve child detection. Our initial results show a 9.2\% absolute improvement over the baseline, leading to a state-of-the-art performance.