4arXiv cs.CL (Computation and Language)·
WordVoice: Word-level acoustic control framework for LLM-based TTS with 4.7k-hour bilingual dataset
Researchers propose WordVoice, a framework enabling explicit, decoupled word-level control over acoustic attributes (duration, boundary, energy, pitch, tone) in LLM-based text-to-speech systems. The work introduces a bound-token mechanism for acoustic planning within the LLM and a fine-grained modulation module to align discrete tokens with continuous waveforms. Accompanying the model is WordVoice-5A, a 4.7k-hour bilingual annotated dataset with five-dimensional word-level labels. The system targets high-precision applications such as audiobook narration and video dubbing where implicit TTS generation falls short.
Related guides (1)
Related events (8)
LeVo 2: Hybrid LLM-Diffusion framework for stable full-length song generation with hierarchical modeling
LeVo 2 is a new hybrid LLM-Diffusion system for controllable full-length song generation that addresses the coherence-vs-acoustics trade-off through hierarchical token prediction: a language model handles semantic planning via mixed tokens, then predicts vocal and accompaniment tracks in parallel, while a diffusion-based codec reconstructs waveforms. A key contribution is an aesthetics-guided progressive post-training schedule combining SFT, offline DPO, and semi-online DPO to separately optimize quality, controllability, and musicality. Expert listening tests show LeVo 2 outperforms open-source baselines across six subjective dimensions and approaches leading commercial systems on several metrics.
UniAudio-Token: Semantic Speech Tokenizer with General Audio Perception for Audio-LLMs
UniAudio-Token is a framework from Tencent that extends semantic speech tokenizers—commonly used as interfaces for Audio-LLMs—to support general audio perception without sacrificing speech quality. It introduces two mechanisms: Semantic-Acoustic Primitives (SAP) for structured supervision decomposing audio into linguistic, vocal, and auditory-scene components, and Semantic-Acoustic Equilibrium (SAE), a content-aware gating mechanism that restores fine-grained acoustic details from shallow layers. Evaluations show it outperforms all single-codebook baseline tokenizers on both understanding and generation tasks when integrated with downstream LLMs. Code, training/inference scripts, and model checkpoints are publicly released.
KIT submission to IWSLT 2026 cross-lingual voice cloning track with language tag prompting and RL fine-tuning
Researchers from KIT describe their system for the IWSLT 2026 Cross-Lingual Voice Cloning shared task, which aims to synthesize speech in a target language while preserving source-speaker identity. The system builds on FishAudio-S2-Pro, a multilingual TTS model, and introduces language tag prompting to reduce accent leakage, RL fine-tuning for intelligibility, and a reference-conditioned lexical matching method for domain-specific pronunciation. Language prompting yields the largest gains; lexical matching provides consistent improvements on matched subsets.
AlignAtt4LLM adapts simultaneous speech translation policy to decoder-only LLMs for IWSLT 2026
Researchers present AlignAtt4LLM, a simultaneous speech translation system for IWSLT 2026 covering English to German, Italian, and Chinese. The system cascades Qwen3-ASR for incremental transcription with Gemma-4 E4B-it for translation, applying a novel AlignAtt policy adapted for decoder-only LLMs that lack encoder-decoder cross-attention. Key contributions include explicit source span prompting, offline alignment head selection, and query/key capture to recover a usable attention-based read/write policy. The system outperforms IWSLT 2026 baselines for European language pairs in both low- and high-latency regimes.
Audio Interaction Model: Unified Streaming LALM with Always-On Perceive-Decide-Respond Loop
Researchers introduce the Audio Interaction Model framework and a concrete implementation called Audio-Interaction, a unified streaming Large Audio Language Model that handles both offline tasks and real-time audio interaction through a continuous perceive-decide-respond loop. The system is built on SoundFlow, a framework covering data construction, training, and asynchronous low-latency inference. The authors also release StreamAudio-2M, a 2.6M-item streaming corpus spanning 28 sub-tasks, and Proactive-Sound-Bench for evaluating proactive audio intervention. Evaluated across 8 benchmarks, the model preserves competitive offline performance while enabling real-time ASR, streaming instruction following, and proactive response capabilities not available in prior offline LALMs.
Multimodal Voice Activity Projection framework for turn-taking prediction in social robots
Researchers introduce MM-VAP, a Multimodal Voice Activity Projection framework that extends audio-only turn-taking prediction to synchronized audio-visual inputs for use in social robots acting as conversation mediators. The approach uses pretrained audio-visual backbones adapted via Low-Rank Adaptation, an inter-speaker attention stage, and a semantic consistency loss to regularize the output space. Experiments on the NoXi, NoXi+J, and Haru EDR corpora show improvements over baselines on several turn-taking events. The work targets human-robot interaction scenarios where anticipating conversational dynamics is required rather than simply reacting to pauses.
BayLing-Duplex: Native full-duplex speech dialogue using a single autoregressive LLM
Researchers introduce BayLing-Duplex, a speech language model that achieves native full-duplex interaction — simultaneous listening and speaking — using a single autoregressive LLM with no auxiliary VAD or turn-taking module. Built by fine-tuning GLM-4-Voice on 400K samples plus a lightweight DPO stage, it reaches 92% turn-taking success and 100% interruption success on InstructS2S-Eval, and improves speech-response quality substantially over Moshi. The approach adds only special tokens to the standard vocabulary, making it portable across LLM architectures without architectural changes.
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.