We present a new 4.1 hours long high-quality motion capture sign language dataset for Swedish Sign Language — STS Mocap v1. The dataset consists of high quality multimodal data: body tracked with markers, fingers tracked with Manus Quantum Metagloves, face tracked with iPhone LiveLink app in MetaHuman Animator mode, and corresponding textual sentence translation to spoken Swedish. With the help of this dataset, we show that four hours of motion capture data is enough for generative modeling of sign language conditioned on 2D pose. In comparison, training the same flow-matching model on only 30 minutes of this data, which is a common size for sign language motion capture datasets, shows a significant degradation in the quality of the synthesized data.

Grounding language in the physical world requires AI systems to interpret references that emerge dynamically during conversation. While current vision-language models (VLMs) excel at static image tasks, they struggle to resolve ambiguous expressions in spontaneous, multi-turn dialogue. We address this gap by introducing MM-Conv—speak, point, look—a benchmark for referential communication in dynamic 3D environments, built from 6.7 hours of egocentric VR interaction with synchronized speech, motion, gaze, and 3D scene geometry. The benchmark includes over 4,200 manually verified referring expressions spanning full, partitive, and pronominal types, enabling systematic evaluation of multimodal reference resolution.

As part of the PSST challenge, we explore how data augmentations, data sources, and model size affect phoneme transcription accuracy on speech produced by individuals with aphasia. We evaluate model performance in terms of feature error rate (FER) and phoneme error rate (PER). We find that data augmentations techniques, such as pitch shift, improve model performance. Additionally, increasing the size of the model decreases FER and PER. Our experiments also show that adding manually-transcribed speech from non-aphasic speakers (TIMIT) improves performance when Room Impulse Response is used to augment the data. The best performing model combines aphasic and non-aphasic data and has a 21.0% PER and a 9.2% FER, a relative improvement of 9.8% compared to the baseline model on the primary outcome measurement. We show that data augmentation, larger model size, and additional non-aphasic data sources can be helpful in improving automatic phoneme recognition models for people with aphasia.

This papers describes a data collection setup and a newly recorded dataset. The main purpose of this dataset is to explore patterns in the focus of visual attention of humans under three different conditions - two humans involved in task-based interaction with a robot; same two humans involved in task-based interaction where the robot is replaced by a third human, and a free three-party human interaction. The dataset contains two parts - 6 sessions with duration of approximately 3 hours and 9 sessions with duration of approximately 4.5 hours. Both parts of the dataset are rich in modalities and recorded data streams - they include the streams of three Kinect v2 devices (color, depth, infrared, body and face data), three high quality audio streams, three high resolution GoPro video streams, touch data for the task-based interactions and the system state of the robot. In addition, the second part of the dataset introduces the data streams from three Tobii Pro Glasses 2 eye trackers. The language of all interactions is English and all data streams are spatially and temporally aligned.

We present an attempt at using 3rd party observer gaze to get a measure of how appropriate each segment in a dialogue is for a speaker change. The method is a step away from the current dependency of speaker turns or talkspurts towards a more general view of speaker changes. We show that 3rd party observers do indeed largely look at the same thing (the speaker), and how this can be captured and utilized to provide insights into human communication. In addition, the results also suggest that there might be differences in the distribution of 3rd party observer gaze depending on how information-rich an utterance is.

We present the Spontal database of spontaneous Swedish dialogues. 120 dialogues of at least 30 minutes each have been captured in high-quality audio, high-resolution video and with a motion capture system. The corpus is currently being processed and annotated, and will be made available for research at the end of the project.