The proliferation of voice assistants for information retrieval is propelled by technological advancements and seamless integration across multiple devices. Nevertheless, these systems face persistent limitations in accuracy and comprehension, particularly with accents, dialects, and uncommon terminology. Additional challenges include the cost of these technologies and their reliance on internet connectivity. This study conducts a comprehensive evaluation of various low-cost speech-to-text transcription software, including Windows10, Google Docs, GBoard Android, Speech-Texter, and SpeechNotes. The analysis focuses on high-error criteria in text retrieval, such as proper names, homophones, neologisms, and multilingual usage. Key variables examined include user age, message duration, and ambient noise levels. Transcription quality is meticulously assessed to determine the efficacy of voice retrieval. Results reveal significant disparities among the software, with GBoard Android demonstrating superior accuracy and the lowest error rates.

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

Comparative Evaluation of Speech-to-Text Software Based on Sociodemographic and Environmental Factors

  • Jorge Morato,
  • Alejandro Pedrero,
  • Sonia Sanchez-Cuadrado

摘要

The proliferation of voice assistants for information retrieval is propelled by technological advancements and seamless integration across multiple devices. Nevertheless, these systems face persistent limitations in accuracy and comprehension, particularly with accents, dialects, and uncommon terminology. Additional challenges include the cost of these technologies and their reliance on internet connectivity. This study conducts a comprehensive evaluation of various low-cost speech-to-text transcription software, including Windows10, Google Docs, GBoard Android, Speech-Texter, and SpeechNotes. The analysis focuses on high-error criteria in text retrieval, such as proper names, homophones, neologisms, and multilingual usage. Key variables examined include user age, message duration, and ambient noise levels. Transcription quality is meticulously assessed to determine the efficacy of voice retrieval. Results reveal significant disparities among the software, with GBoard Android demonstrating superior accuracy and the lowest error rates.