InaGVAD : a Challenging French TV and Radio Corpus Annotated for Speech Activity Detection and Speaker Gender Segmentation
David Doukhan, Christine Maertens, William Le Personnic, Ludovic Speroni, Reda Dehak
TL;DR
InaGVAD addresses the need for a diverse, richly annotated audiovisual resource to study speech activity detection and speaker gender segmentation in French media. The authors compile 277 one-minute excerpts from 28 channels across four categories, with fine-grained annotations for speech, non-speech events, overlaps, and speaker traits, and split the data into a 1-hour development set and a 3-hour-37-minute test set to enable fair benchmarking. They benchmark six open-source VAD systems and three SGS approaches, including a transfer-learning baseline using ECAPA-TDNN embeddings, showing channel-dependent performance and highlighting the particular difficulty of generalist TV and music radio content. The study demonstrates that a lightweight, diverse development set can yield competitive SGS results and provides a fully open pipeline (data, annotations, code) to support reproducible research, encouraging future improvements in audiovisual VAD/SGS methods and analysis of gender representation in media.
Abstract
InaGVAD is an audio corpus collected from 10 French radio and 18 TV channels categorized into 4 groups: generalist radio, music radio, news TV, and generalist TV. It contains 277 1-minute-long annotated recordings aimed at representing the acoustic diversity of French audiovisual programs and was primarily designed to build systems able to monitor men's and women's speaking time in media. inaGVAD is provided with Voice Activity Detection (VAD) and Speaker Gender Segmentation (SGS) annotations extended with overlap, speaker traits (gender, age, voice quality), and 10 non-speech event categories. Annotation distributions are detailed for each channel category. This dataset is partitioned into a 1h development and a 3h37 test subset, allowing fair and reproducible system evaluation. A benchmark of 6 freely available VAD software is presented, showing diverse abilities based on channel and non-speech event categories. Two existing SGS systems are evaluated on the corpus and compared against a baseline X-vector transfer learning strategy, trained on the development subset. Results demonstrate that our proposal, trained on a single - but diverse - hour of data, achieved competitive SGS results. The entire inaGVAD package; including corpus, annotations, evaluation scripts, and baseline training code; is made freely accessible, fostering future advancement in the domain.