Saved in:
Bibliographic Details
Main Authors: Mohapatra, Payal, Likhite, Shamika, Biswas, Subrata, Islam, Bashima, Zhu, Qi
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.06964
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909221067423744
author Mohapatra, Payal
Likhite, Shamika
Biswas, Subrata
Islam, Bashima
Zhu, Qi
author_facet Mohapatra, Payal
Likhite, Shamika
Biswas, Subrata
Islam, Bashima
Zhu, Qi
contents Most existing speech disfluency detection techniques only rely upon acoustic data. In this work, we present a practical multimodal disfluency detection approach that leverages available video data together with audio. We curate an audiovisual dataset and propose a novel fusion technique with unified weight-sharing modality-agnostic encoders to learn the temporal and semantic context. Our resilient design accommodates real-world scenarios where the video modality may sometimes be missing during inference. We also present alternative fusion strategies when both modalities are assured to be complete. In experiments across five disfluency-detection tasks, our unified multimodal approach significantly outperforms Audio-only unimodal methods, yielding an average absolute improvement of 10% (i.e., 10 percentage point increase) when both video and audio modalities are always available, and 7% even when video modality is missing in half of the samples.
format Preprint
id arxiv_https___arxiv_org_abs_2406_06964
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Missingness-resilient Video-enhanced Multimodal Disfluency Detection
Mohapatra, Payal
Likhite, Shamika
Biswas, Subrata
Islam, Bashima
Zhu, Qi
Computation and Language
Multimedia
Sound
Audio and Speech Processing
Most existing speech disfluency detection techniques only rely upon acoustic data. In this work, we present a practical multimodal disfluency detection approach that leverages available video data together with audio. We curate an audiovisual dataset and propose a novel fusion technique with unified weight-sharing modality-agnostic encoders to learn the temporal and semantic context. Our resilient design accommodates real-world scenarios where the video modality may sometimes be missing during inference. We also present alternative fusion strategies when both modalities are assured to be complete. In experiments across five disfluency-detection tasks, our unified multimodal approach significantly outperforms Audio-only unimodal methods, yielding an average absolute improvement of 10% (i.e., 10 percentage point increase) when both video and audio modalities are always available, and 7% even when video modality is missing in half of the samples.
title Missingness-resilient Video-enhanced Multimodal Disfluency Detection
topic Computation and Language
Multimedia
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2406.06964