VieSpeaker: A Large-Scale Vietnamese Speaker Recognition Dataset Beyond Visual Dependency

A research team has introduced VieSpeaker, a Vietnamese speaker recognition dataset containing roughly 902 hours of speech from 4,715 speakers, built without relying on facial cues to link voices to identities [1][2]. The dataset, detailed in a paper posted to the arXiv preprint repository on June 23, 2026, addresses a gap in Vietnamese speech resources, which the authors describe as limited in scale and acoustic diversity [1][2]. Most large-scale speaker recognition corpora depend on video recordings where speakers appear on camera, a constraint that narrows the range of usable data [2]. The VieSpeaker pipeline instead uses textual metadata and large language model reasoning to infer who is speaking from transcripts and surrounding context [1][2]. Large language models are machine learning systems trained on vast amounts of text to perform language tasks such as generation and inference [8]. The paper reports that models trained on VieSpeaker show improved robustness and generalization when compared with existing Vietnamese datasets [1][2]. The authors frame the work as a demonstration that face-independent dataset construction is feasible and offers a new direction for building large-scale speech resources [1][2]. The paper appears on arXiv, an open-access repository of electronic preprints that is moderated but not peer-reviewed [6]. Founded in 1991, arXiv passed two million articles by the end of 2021 and now receives about 24,000 submissions per month [6]. The VieSpeaker preprint is listed under the Computer Science and Sound categories and is accompanied by experimental tools available through arXivLabs, a framework that lets community collaborators develop and share features on the site [1][4]. arXivLabs projects, which include citation explorers and code-finding tools, operate under guidelines that require partners to uphold openness, community, excellence, and user data privacy [4][5].