speech-processing

The paper introduces DuplexSpeechBench-IFEval, a benchmark for evaluating implicit instruction following in full-duplex voice agents, with 1,038 test cases across eight roles and five protocols to assess real-time speech systems' adherence to explicit vs. persona-implied behaviors.

AVERT enhances spoken dialogue state tracking by integrating audio verification with cross-turn agreement to fix errors, boosting performance on the SpokenWOZ benchmark without retraining.

The paper introduces VoiceLongMemEval (VLME), a benchmark that evaluates AI assistants' ability to remember and reason over paralinguistic metadata like emotion and prosody from voice in long-term conversations, revealing an 'affect gap' in current models.

This paper presents V2TATC, a joint voice–trajectory embedding framework for air traffic controller situational awareness, and introduces a novel dataset for cross-modal retrieval experiments in congested airspaces.

This paper proposes a generalized style-aware full-duplex framework with a lightweight turn controller LPS-TC, introduces a large-scale dataset WildTurn for real-world conversations, and presents a two-tier evaluation scheme to enhance proactive spoken interactions and response quality in dialogue systems.

This paper proposes STeReO, a reranker for orchestrating heterogeneous speech and text retrievers to enhance Retrieval-Augmented Generation systems by improving evidence selection in multi-modality scenarios.

The study tests real-time voice AI models on emotional cues in speech, finding that while they can detect emotions like distress or sarcasm, this information doesn't reliably influence decisions, highlighting an emotional intelligence gap in current systems.

NVIDIA has launched Nemotron 3 VoiceChat, an AI model enabling real-time, full-duplex voice interactions that allow natural interruptions, available via a live demo as part of its open-source NeMo Speech framework.

A tweet shares a 17-page PDF from Stanford University that explains Hidden Markov Models, a foundational concept in machine learning and speech processing.

This paper introduces a novel training criterion to reduce reliance on shortcuts in automated spoken English proficiency assessment systems, aiming to prevent test-takers from inflating scores without genuine improvement. Experiments on audio and text-based systems show reduced correlation with exploitable features, bringing automated scores closer to human judgment.

This paper presents SAMPA, a Whisper-based segmenter fine-tuned on Brazilian Portuguese speech data to automatically mark terminal prosodic boundaries, achieving competitive F1 scores on held-out and out-of-distribution datasets.

The paper proposes a multimodal solution for audio sentiment polarity classification that integrates audio and multilingual text transcripts via cross-modal transformers, and uses knowledge distillation to enhance an audio-only model without computational overhead during inference.

This paper investigates the use of conversational temporal dynamics (turn-pair timing) as a lightweight modality for automatic depression detection from dyadic clinical interviews, showing that a compact 24-dimensional timing module achieves strong performance and complements standard acoustic and semantic features when fused.

This paper describes NAVER LABS Europe's submission to the IWSLT 2026 instruction-following short track, improving upon their previous winning system by using a new speech projector (SpeechMapper) trained solely on ASR data and augmenting training with a synthetic SQA dataset (fakACL). The resulting system ties for first place in the constrained track while using a weaker LLM backbone.

A survey of automated presentation coaching systems that reviews existing systems, introduces a five-dimensional task taxonomy covering pronunciation, stress, prosody, pacing, and content faithfulness, and identifies open challenges such as annotation scarcity, accent fairness, and low-latency feedback.

The Jan 6, 2026 draft of the 3rd edition of 'Speech and Language Processing' by Dan Jurafsky and James H. Martin is released, featuring a revised structure with a focus on large language models and updated chapters.

A C++ implementation of distilHuBERT with no runtime dependencies, compiled-in weights, dynamic sizing, and on-par performance with ONNX Runtime, designed for easy integration into CMake projects.

This paper introduces a semantic-timescale analysis pipeline to study how generic vs. specific content is distributed over time in human and AI-generated speech, revealing that autocorrelation-window measures capture temporal organization of semantics beyond static lexical distributions.

A novel method for multilingual word-level forced alignment combines self-supervised representations from MMS and a phoneme boundary detector with a learned dynamic programming decoder, outperforming existing aligners on English and unseen languages without further training.

InfoShield introduces a privacy-preserving method for speech representations in mental health screening using information-theoretic optimization, reducing sensitive attribute inference while maintaining diagnostic accuracy. A novel TimeAwareMINE estimator addresses temporal-static misalignment in sequential speech.