Dataset of audiovisual unscripted monologues with speech annotations

This article presents a publicly accessible multimodal dataset comprising audiovisual recordings of unscripted German monologues and a corresponding time-resolved linguistic speech annotation corpus. The audiovisual dataset consists of more than 300 minutes of natural speech material recorded from six speakers across 60 takes. Recordings were captured using a Canon EOS 700D camera and a Neumann KM-184 microphone, with audio sampled at 48 kHz and video at 25 frames per second. Speakers produced spontaneous monologues on self-selected everyday topics under controlled laboratory conditions. Recording sessions included variations such as background babble noise presented via in-ear headphones and optional visual modifications, including the use of glasses or lipstick. Audio and video were acquired simultaneously, synchronized during post-processing, and segmented into individual stories. The accompanying speech annotation corpus provides detailed linguistic information aligned with the audiovisual material at millisecond resolution. An annotation pipeline combining established tools-OCTRA, G2P, MAUS, PHO2SYL, and the RFTagger-was used to derive orthographic transcripts, canonical phonological representations, phonetic segmentations, syllabification, and part-of-speech tags. Manual correction steps were applied to ensure transcription accuracy and to improve the quality of forced alignment. Each recording is accompanied by an events.tsv file containing time-stamped word, phoneme, and pause annotations; a JSON sidecar describing variable metadata; and a machine-actionable HED-formatted event file to support integration with neuroimaging standards such as BIDS. The structure of the resource follows a consistent file-naming scheme to ensure reliable linkage between audiovisual recordings and speech annotations. This combined audiovisual and linguistic resource supports a wide range of reuse applications, including acoustic-phonetic analysis, linguistic and neurolinguistic research, annotation benchmarking, and the development or evaluation of speech-processing tools. The resource's naturalistic content and high temporal precision enable detailed examination of spontaneous speech and facilitate replication-oriented research across disciplines.