Tag
本研究首次探索从韵律模式反向预测文本：通过仅使用最低 12 个 Mel 频带（约 450Hz 截止的低通滤波）微调 Whisper 模型，实现了 36% 的词错误率，表明低频语音特征与词汇内容的关联远强于此前认知。
This study examines charismatic prosody in bilingual political speeches across Luxembourgish and French, finding that speaker identity accounts for most variance while language shows systematic differences in acoustic features.
The paper introduces Hear2Act, a unified benchmark for evaluating how prosodic cues affect task-oriented dialogue decisions, finding that prosody matters when lexical evidence is insufficient and that audio LLMs benefit from explicit concern representation for actions.
This study investigates whether contextualized embeddings from a large language model predict spoken word duration and pitch contours for Mandarin monosyllabic words, demonstrating above-chance prediction and the ability to back-transform normalized f0 contours to ms time scale.
This paper proposes a reference-based evaluation protocol for assessing prosody and rhythm in speech-to-speech AI systems, using matched human conversation data to provide interpretable behavioral plausibility checks.
Introduces MMEE, a multilingual multi-emotion emphasis corpus of 10,000 utterances across 7 languages and 34 emotions, and benchmarks emphasis detection models under various transfer settings, finding that multilingual training improves robustness while monolingual models show limited zero-shot transfer.
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.
Proposes TextPro-SLM, a speech large language model that minimizes the modality gap by processing spoken input to resemble prosody-aware text input, achieving strong paralinguistic understanding with low training data.