Low-Resource Contextual Topic Identification on Speech

TL;DR

This paper introduces a novel attention-based contextual model for low-resource spoken topic identification, significantly improving accuracy by leveraging sequential segment dependencies in unstructured audio.

Contribution

It presents a new attention-based approach that exploits contextual information across segments, enhancing topic ID performance in low-resource language settings.

Findings

Attention-based model outperforms non-contextual models on four languages

Contextual dependencies significantly improve topic classification accuracy

Method is effective for low-resource spoken language processing

Abstract

In topic identification (topic ID) on real-world unstructured audio, an audio instance of variable topic shifts is first broken into sequential segments, and each segment is independently classified. We first present a general purpose method for topic ID on spoken segments in low-resource languages, using a cascade of universal acoustic modeling, translation lexicons to English, and English-language topic classification. Next, instead of classifying each segment independently, we demonstrate that exploring the contextual dependencies across sequential segments can provide large improvements. In particular, we propose an attention-based contextual model which is able to leverage the contexts in a selective manner. We test both our contextual and non-contextual models on four LORELEI languages, and on all but one our attention-based contextual model significantly outperforms the context-independent models.