Scholarlink

<p>The multi-disciplinary field of voice perception and trustworthiness lacks accessible and diverse speech audio datasets representing diverse speaker demographics, including age, ethnicity, and sex. Existing datasets primarily feature white, younger adult speakers, limiting generalisability. This paper introduces a novel open-access speech audio dataset with 1,152 utterances from 96 untrained speakers, across white, black and south Asian backgrounds, divided into younger (N = 60, ages 18–45) and older (N = 36, ages 60+) adults. Each speaker recorded both, their natural speech patterns (i.e. “neutral” or no intent), and their attempt to convey their trustworthy intent as they perceive it during speech production. Our dataset is described and evaluated through classification methods between neutral and trustworthy speech. Specifically, extracted acoustic and voice quality features were analysed using linear and non-linear classification models, achieving accuracies of around 70%. This dataset aims to close a crucial gap in the existing literature and provide additional research opportunities that can contribute to the generalisability and applicability of future research results in this field.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Human voices communicating trustworthy intent: A demographically diverse speech audio dataset

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月31日

查看全文
  • Constantina Maltezou-Papastylianou,
  • Reinhold Scherer,
  • Silke Paulmann

摘要

The multi-disciplinary field of voice perception and trustworthiness lacks accessible and diverse speech audio datasets representing diverse speaker demographics, including age, ethnicity, and sex. Existing datasets primarily feature white, younger adult speakers, limiting generalisability. This paper introduces a novel open-access speech audio dataset with 1,152 utterances from 96 untrained speakers, across white, black and south Asian backgrounds, divided into younger (N = 60, ages 18–45) and older (N = 36, ages 60+) adults. Each speaker recorded both, their natural speech patterns (i.e. “neutral” or no intent), and their attempt to convey their trustworthy intent as they perceive it during speech production. Our dataset is described and evaluated through classification methods between neutral and trustworthy speech. Specifically, extracted acoustic and voice quality features were analysed using linear and non-linear classification models, achieving accuracies of around 70%. This dataset aims to close a crucial gap in the existing literature and provide additional research opportunities that can contribute to the generalisability and applicability of future research results in this field.