Assessing automatic VOT annotation using unimpaired and impaired speech

Area of Science:

• Phonetics
• Speech Science
• Computational Linguistics

Background:

• Manual phonetic annotation of speech recordings is crucial for studying speech processes but is labor-intensive and costly.
• Advances in automatic annotation offer potential cost reductions, but their validity compared to manual methods needs evaluation.

Purpose of the Study:

• To assess the accuracy and reliability of the AutoVOT automatic speech annotation tool in approximating manual phonetic annotation.
• To determine if research conclusions about speech processes remain consistent when using automatically annotated data versus manually annotated data.

Main Methods:

• Comparison of analyses derived from speech data annotated by the AutoVOT tool against analyses from manually annotated speech.
• Evaluation conducted across two distinct studies to ensure robustness of findings.

Main Results:

• AutoVOT demonstrated a high degree of approximation to manual annotation.
• Similar conclusions regarding speech processes were achievable using both automatic and manual annotation methods, with specific considerations.

Conclusions:

• Automatic phonetic annotation, exemplified by AutoVOT, presents a viable alternative to manual annotation for reducing costs in speech research.
• Guidelines are provided for researchers and clinicians on the appropriate application and potential limitations of AutoVOT in various datasets.