핵심 개념
Crowdsourced multilingual speech intelligibility testing offers a cost-efficient and scalable approach to assess speech quality across languages.
초록
Introduction: Discusses the need for rapid evaluation of generative audio features on speech intelligibility.
Objective Measures: Objective measures are crucial for algorithm development, especially with generative approaches.
Listening Tests: Traditional listening tests in laboratories are costly and time-consuming, hindering rapid algorithm research.
Crowdsourced Assessments: Crowdsourcing provides a scalable and cost-effective alternative for assessing speech quality.
Speech Intelligibility Assessment: Various methods like DRT and MRT are used for assessing speech intelligibility.
Crowdsourcing Test Design: Details the implementation of DRT as a repeatable online survey using Qualtrics.
Experiments and Results: Presents experiments on Spanish accuracy, consistency, codec comparisons, and wideband vs. PCMU across languages.
Discussion and Future Work: Highlights the benefits of crowdsourced assessments and future research directions.
통계
"The NB PCMU score was 4.3 points below the WB reference score."
"Participants received rewards corresponding to an hourly rate exceeding $8.00 USD on average."
인용구
"Listening tests remain the gold standard for assessing quality and intelligibility of speech."
"Crowdworkers scored 2.6 points lower than expert listeners in the NB PCMU condition."