Scholarlink

<p>Voice-based artificial intelligence (AI) has become deeply embedded in everyday digital interaction, mediating routine activities such as message composition, reminder setting, service navigation, and the control of domestic environments across both personal and professional contexts. Despite the apparent fluency of these systems, instances of failure reveal important limitations in their underlying architectures. This paper identifies a convergence not previously examined: English people with speech impairments and speakers of Persian, two populations with no shared linguistic or physiological profiles that trigger structurally identical failure patterns in ASR systems, including hallucination and language confusion. The review traces both cases to a shared mechanism: distributional exclusion. In the cases examined here, communication breakdowns cannot be attributed to user input: in both instances, the speakers conveyed their intentions clearly; rather, the failure occurred at the level of the model’s interpretative processing.</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

When AI fails to listen: convergent failure patterns across speech impairments and low-resource speech in ASR

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Reyhaneh Sohrabi

摘要

Voice-based artificial intelligence (AI) has become deeply embedded in everyday digital interaction, mediating routine activities such as message composition, reminder setting, service navigation, and the control of domestic environments across both personal and professional contexts. Despite the apparent fluency of these systems, instances of failure reveal important limitations in their underlying architectures. This paper identifies a convergence not previously examined: English people with speech impairments and speakers of Persian, two populations with no shared linguistic or physiological profiles that trigger structurally identical failure patterns in ASR systems, including hallucination and language confusion. The review traces both cases to a shared mechanism: distributional exclusion. In the cases examined here, communication breakdowns cannot be attributed to user input: in both instances, the speakers conveyed their intentions clearly; rather, the failure occurred at the level of the model’s interpretative processing.