3arXiv cs.CL (Computation and Language)·
Study compares silence/turn-taking gaps in human sitcoms vs. Google NotebookLM synthetic podcasts
A preprint investigates silence gaps and turn-taking patterns in two corpora: thirty US sitcoms and fifty-one synthetic podcasts generated by Google NotebookLM. The study compares gap durations across speaker gender (estimated via fundamental frequency in Praat) and production settings. The work provides empirical data on how AI-generated conversational audio differs from human discourse in prosodic timing.
Related guides (2)
Related events (8)
ParaPairAudioBench: Pairwise benchmark reveals large gaps in LALM paralinguistic judgment
Researchers introduce ParaPairAudioBench, a pairwise audio benchmark of 5,175 audio pairs spanning five paralinguistic dimensions (Style, Rate, Emphasis, Age, Gender) designed to evaluate Large Audio-Language Models as judges. Experiments show current LALMs lag human judgment by 32 percentage points on average and exhibit severe calibration failures, especially in ambiguous 'Tie' cases. The benchmark includes same-transcript and cross-transcript conditions to disentangle lexical from acoustic reliance, enabling more rigorous assessment of LALM reliability for speech evaluation.
SPEARBench: Benchmark for naturalness evaluation in streaming speech-to-speech language models
Researchers introduce SPEARBench, a benchmark for evaluating conversational naturalness in streaming speech-to-speech language models, covering dimensions such as response latency, turn-taking, prosody, dialect consistency, emotional adaptation, and interpersonal stance. The benchmark constructs controlled dialogue prompts from the Seamless Interaction corpus and evaluates multiple contemporary models against human reference answers. Results reveal that current models achieve high signal-level quality and low ASR error rates but still diverge from human conversational behavior on latency, overlap, dialect preservation, and stance dynamics. The work addresses a gap in existing speech and text benchmarks that fail to capture conversational naturalness.
Benchmarking human vs. ASR systems on diverse Dutch speech (children, elderly, Flemish)
A preprint from arXiv compares state-of-the-art ASR systems against Dutch native listeners on 'diverse' speech including child speech, older adult speech, and Flemish regional accents. Google Telephony outperformed other ASR systems, and ASR systems matched or exceeded human listener performance in several conditions. Performance gaps remain for aging-related acoustic variability and regional accents, and the choice of test set materially affects benchmark conclusions.
Instruction-tuned LLMs show stronger syntactic convergence toward human interlocutors than humans do
A new arXiv preprint measures syntactic convergence in 16 open-weight Llama and Gemma models (1B–70B, pretrained and instruction-tuned) by substituting model outputs into pre-existing human dialogues and measuring context-free grammar rule reuse. Every model showed greater CFG-rule overlap with the preceding human turn than a random baseline, and instruction-tuned models exceeded the syntactic convergence of the human responses they replaced. Instruction tuning increased local syntactic and semantic alignment with interlocutors, though it also raised baseline overlap with unrelated primes, suggesting a more globally accommodating output style rather than purely targeted priming.
Synthetic LLM-generated conversations improve ASR training for low-resource languages
Researchers propose a pipeline that uses LLMs to generate scenario-level dialogues and TTS to synthesize multi-speaker audio, creating simulated conversational training data for ASR systems. Evaluated on the Hungarian BEA-Dialogue benchmark, a model trained on 67 hours of real plus 636 hours of synthetic data outperforms a zero-shot model trained on 2,700 hours of real Hungarian speech. The study tests five LLM families under multiple budget and mixing configurations using a FastConformer-Large backbone, finding that generator choice and data composition significantly affect gains.
ModeratorLM: Role-conditioned turn-taking for multi-party voice agents with 40%+ precision gains
Researchers introduce ModeratorLM, a voice agent system that conditions turn-taking behavior on an explicitly assigned conversational role in multi-party settings, built on a streaming speech LLM. A reasoning-augmented variant adds chain-of-thought over conversational context. Evaluated on real-world meeting data and the new RolePlayConv synthetic dataset, the system achieves over 40% improvement in turn-taking precision and 70% in recall while reducing false-positive interruptions versus non-role-conditioned baselines.
Study finds lower bitrate discrete speech representations sufficient for generative spoken language modeling
Researchers investigate how segmentation width and cluster size affect speech resynthesis and continuation quality in Generative Spoken Language Models (GSLM), which train language models on discrete speech units without text. They find that intelligible, natural speech can be synthesized at lower bitrates than the standard baseline, and that continuation quality remains stable at reduced bitrates, suggesting conventional GSLM settings may be over-specified. The paper also notes that LLM-based evaluation metrics correlate better with human judgments than conventional metrics, but correlation remains low, pointing to a gap in automatic evaluation for speech generation.
Audit reveals protocol-level shortcuts in large audio-language model judges for speech evaluation
Researchers audit 'protocol-level shortcuts' in large audio-language models (LALMs) used as automatic judges for speech evaluation, testing across three deployment protocols: feature-blueprint judging, reference-conditioned judging, and pairwise A/B comparison. Across six judges and four attributes, several LALMs are found to rely on shortcuts rather than actual audio content — for example, incorrect specialist labels collapse emotion accuracy to 0.10 or below for five judges, and Qwen3-Omni-Thinking shows position bias in A/B comparisons. The findings indicate that high human-agreement scores can overstate judge validity, and that each model-protocol pair requires its own shortcut probe to be trustworthy.