OSynth is a large-scale synthetic dataset that simulates organizations, providing 570 hours of meeting audio along with related artifacts and a latent ground-truth state. Human raters can only distinguish OSynth meetings from real transcripts 56% of the time when given short excerpts, indicating the dataset’s realism. When tested with state-of-the-art large language models, adding contextual information from OSynth significantly improves performance on entity-sensitive tasks, highlighting both current limitations and the dataset’s potential for advancing contextual meeting understanding.
arXiv:2604.17260v3 Announce Type: replace Abstract: Evaluating meeting effectiveness is crucial for improving organizational productivity. Current approaches rely on post-hoc surveys that yield a sin...
arXiv:2609.39344v1 Announce Type: cross Abstract: Speech transcripts used as long-term memory must preserve both words and stable speaker identities. Existing meeting-transcription metrics either ign...
Speech transcripts used as long-term memory must preserve both words and stable speaker identities. Existing meeting-transcription metrics either ignore speakers or remap anonymous speakers independen...
arXiv:2609.22452v1 Announce Type: new Abstract: Long-context understanding remains a fundamental challenge for large language models, as excessively long inputs often lead models to forget salient in...
The paper introduces an LLM-assisted system for correcting speaker attribution errors during meetings. It combines streaming ASR, diarization, and concise LLM-generated summaries to guide users in providing brief corrective feedback, which updates the transcript and adds online speaker enrollments. The approach includes mechanisms to accurately interpret user corrections and a simulation for large-scale evaluation, achieving significant reductions in DER and speaker substitution error on the AMI headset test set, with a pilot usability study highlighting further improvements.
The paper introduces LOGIC (Logit‑Space Integration for Contextual Biasing), a new framework that injects contextual entity information directly into the decoding layer of Speech Large Language Models, bypassing the limitations of prompt‑based methods. LOGIC operates with constant‑time complexity regardless of the size of the entity list, and experiments with the Phi‑4‑MM model across 11 multilingual locales show an average 9% relative reduction in Entity WER while adding only a 0.30% increase in False Alarm Rate.
The paper introduces a synthetic data generation pipeline for long‑context audio reasoning, focusing on first‑visit doctor‑patient conversations and SOAP note generation. The pipeline consists of persona‑driven dialogue creation, multi‑speaker audio synthesis with realistic acoustics, and LLM‑based reference note production, all built with open‑weight models. The authors release 8,800 synthetic conversations, 1.3k hours of audio, and reference notes, and show that cascaded approaches still outperform end‑to‑end models on current open‑weight systems.
arXiv:2609.26780v1 Announce Type: cross Abstract: Long-term conversational memory in multi-party settings requires more than retrieving relevant content from long-term conversations: it must distingu...
The paper introduces a pipeline that generates intent‑labeled, two‑channel conversational speech from relational event lists, enabling controlled synthesis of full‑duplex dialogue with 42 phenomena across eight families in English and Mandarin. By having an LLM author each event’s speaker, text, conversational act, and attachment, and then aligning and timing these events independently, the system produces diverse, realistic turn‑taking signals. Experiments show that models trained on this synthetic corpus achieve higher floor‑occupancy accuracy and better start‑speaking/listening F1 scores compared to models trained on prior data.
arXiv:2610.07338v1 Announce Type: cross Abstract: Audio benchmarks are built around short, pre-segmented clips, limiting model design to brief inputs or fixed vocabularies. To close this gap, we intr...
arXiv:2607. 03744v1 Announce Type: new Abstract: Automatic depression detection from clinical interviews typically models the semantic content and acoustic characteristics of participant speech.
PERSIST is a persistent memory system designed for multi-session, multi-speaker spoken dialogue, explicitly modeling Who, What, and When to enable voice assistants to recall past conversations and adapt to individual users. It structures cross-session histories into readable event records and retrieves them using a 3W joint scoring mechanism that combines semantic content, acoustic speaker identity, and temporal state. By reusing intermediate representations from the dialogue backbone, PERSIST reduces retrieval latency from 578.42 ms to 7.03 ms and achieves 85.08% end‑to‑end task accuracy on the SpokenTrace benchmark, improving EM@3 from 49.01% to 82.10%.