Tag
AutoTerm-SST 是一个交互式流式语音翻译系统，通过自适应术语记忆和话题切片路由，实现零会话设置的术语感知翻译，无需用户上传术语表或选择领域即可匹配已知领域基线的术语准确率与 BLEU 表现。
PEDBench is the first phonetic error detection benchmark spanning natural, acted, and synthetic speech, evaluating 13 speech models and finding they fail to reliably detect phonetic errors. Fine-tuned speech-language models on synthetic data outperform baselines on word-level detection, though a sim-to-real gap remains.
提出一种通过LLM生成方言文本并用标准语TTS合成伪方言语音的零真实方言数据增强方法，并引入中间标准文本预测以缩小语义差距，提升语音语言模型在日语、德语和中文方言上的翻译表现。
This paper adapts a full-utterance speech language model for simultaneous speech translation using self-supervised prefix translations, showing that multi-turn append-only decoding with confidence-based commit control improves quality-latency frontiers and sharply reduces commit-calibration error on FLEURS and CoVoST 2.
This paper introduces HEAR, a benchmark for evaluating speaker-attributed reasoning in speech language models, and presents A2R, a 30B model optimized with counterfactual data to improve performance on multi-speaker tasks.
VoiceMem introduces a dual-brain streaming memory architecture for speech language models that improves retrieval accuracy, emotional personalization, and real-time efficiency.
This preprint introduces a generation-aligned diagnostic ladder that separates decision-rule misalignment from readout-coverage limitations in speech language models, showing that state decoding far exceeds generated accuracy in emotion recognition tasks.
This paper reveals that interleaved speech-text language models implicitly transcribe speech into text in intermediate layers, then predict in text space before converting back to speech, shedding light on internal modality interaction.
ParaBridge is an on-policy self-distillation method that bridges the gap between paralinguistic perception and dialogue behavior in speech language models, significantly improving safety and empathy without external rewards.
This paper investigates whether factual recall mechanisms learned in text-based language models transfer to speech modalities in multimodal speech-language models. Using causal mediation analysis on SpiritLM, it finds that the mechanisms are only partially carried over, highlighting differences between text and speech processing.
The Awesome-SpeechLM-Survey repository on GitHub systematically organizes the research lineage of speech language models, including classification frameworks, representative models, training datasets, and evaluation benchmarks. It serves as a knowledge map for understanding the field.
MoshiRAG combines a compact full-duplex speech language model with asynchronous retrieval-augmented generation to improve factuality while maintaining real-time interactivity. The approach leverages natural temporal gaps in conversation to retrieve external knowledge without disrupting the natural flow of dialogue.
MTR-DuplexBench introduces a comprehensive benchmark for evaluating Full-Duplex Speech Language Models in multi-round conversations, addressing challenges like blurred turn boundaries and context inconsistency while assessing conversational features, dialogue quality, instruction following, and safety.