wav2VOT: Automatic estimation of voice onset time, closure duration, and burst realisation with wav2vec2

A new tool called wav2VOT uses the wav2vec2 speech model to automatically estimate voice onset time, closure duration, and burst realisation, according to research posted on arXiv [1]. The authors report the system performs comparably with current methods on unseen data and achieves high accuracy after fine-tuning [2]. The work addresses a persistent bottleneck in phonetics research, where automated speech annotation tools often require substantial manual correction or large training sets to reach acceptable accuracy [2]. The researchers behind wav2VOT set out to test whether large speech models, already proven effective at speech classification tasks, could be repurposed for detailed phonetic annotation [2]. Their analysis shows the tool's predictions maintain high fidelity across stop voicing and place of articulation [2]. The paper was submitted to arXiv on 27 June 2026 under the Sound category within computer science [1]. arXiv, which began on 14 August 1991, serves as an open-access repository for electronic preprints and postprints that are moderated but not peer reviewed [6]. The platform passed the two-million-article milestone by the end of 2021 and now receives roughly 24,000 submissions per month [6]. The wav2VOT paper appears alongside experimental community tools developed through arXivLabs, a framework launched in 2020 that allows third-party collaborators to build features on the repository's article pages [5]. arXivLabs projects, which include bibliographic explorers and code-finding tools, operate under guidelines requiring partners to share arXiv's values of openness, community, excellence, and user data privacy [5]. The arXiv team is currently pausing new Labs proposals while it focuses on modernizing and migrating its systems to the cloud, though existing projects and already-submitted proposals are unaffected [3]. The authors of wav2VOT conclude that their results demonstrate large speech models can produce accurate annotations and warrant further exploration of such models as tools in phonetic research pipelines [2].