The paper investigates how different speech content representations—such as SSL features, supervised tokens, posteriorgrams, and neural audio codecs—perform when used to train a generative model that produces audio conditioned only on each representation. By evaluating the generated audio on content, speaker identity, and prosody, the study identifies two regimes: some representations almost fully reconstruct the original audio, while others effectively separate speaker identity. The findings reveal that disentanglement of speaker identity depends on both the training objective and the representation’s information capacity, rather than supervision alone.
By Diego Torres, Axel Roebel, Nicolas Obin
arXiv:2609.22851v2 Announce Type: cross
Abstract: Large Audio Language Models (LALMs) utilize either continuous features or discrete tokens, yet the optimal representation paradigm for general audio...
By Jing Peng, Zichao Nie, Zhisheng Zhang, Jingran Xie, Zhiyong Wu
arXiv:2605.27840v2 Announce Type: replace-cross
Abstract: Audio tokenizers are fundamental to unifying audio understanding and generation. Understanding requires high-level semantics, while generatio...
By Zhisheng Zhang, Xiang Li, Yixuan Zhou, Jing Peng, Guoyang Zeng, Zhiyong Wu
arXiv:2601. 09239v5 Announce Type: replace-cross Abstract: Speech tokenizers are a key building block of fully discrete Speech LLMs.
By Hanlin Zhang, Daxin Tan, Dehua Tao, Xiao Chen, Haochen Tan, Yunhe Li, Yuchen Cao, Linqi Song
arXiv:2607. 29363v1 Announce Type: cross Abstract: Balancing sequence length, representational capacity, and long-horizon stability is a central problem in autoregressive (AR) speech and audio generation.
By Yi Luo, Rongzhi Gu, Jixun Yao
arXiv:2605. 29948v2 Announce Type: replace-cross Abstract: Unified speech foundation models require a holistic tokenization space that is both learnable by language models and decodable into high-quality waveforms.
By Bohan Li, Shi Lian, Hankun Wang, Yiwei Guo, Yu Xi, Zhihan Li, Da Zheng, Colin Zhang, Kai Yu
FireRedAudio is a 9‑billion‑parameter audio language model that separates continuous input representations for audio understanding and speech generation, enabling a single autoregressive LLM to perform tasks such as ASR, zero‑shot TTS, Instruct TTS, and semantic/acoustic speech editing. The model uses a dedicated Audio Encoder for recognition and a RedAE‑based pathway for generation, with the LLM directly generating text or conditioning a flow‑matching DiT to produce acoustic latents. Evaluations show competitive or leading performance in multilingual ASR, content‑accurate zero‑shot TTS, strong instruction following, and significant improvements in speech editing over prior work.
By Feiyu Shen, Fenglong Xie, Junjie Li, Kun Xie, Lei Xie, Xu Tang, Xuelong Geng, Yan Jia, Yao Hu, Yichen Han, Yichen Wu, Ziqi Dai, Junjie Chen, Kai Huang, Manzhen Wei, Yixuan Li
arXiv:2605.29948v3 Announce Type: replace-cross
Abstract: Unified speech foundation models require a holistic tokenization space that is both learnable by language models and decodable into high-qual...
By Bohan Li, Shi Lian, Hankun Wang, Yiwei Guo, Yu Xi, Zhihan Li, Da Zheng, Colin Zhang, Kai Yu
Neural audio codecs are a key component of speech processing pipelines, compressing audio into discrete tokens for downstream modeling. However, existing codecs struggle to balance reconstruction quality with token efficiency, often encoding perceptually irrelevant information such as background noise and recording artifacts at the expense of linguistically and acoustically meaningful content.
Audio intelligence involves understanding, reasoning about, and generating both audio and speech. In this work, we introduce Nemotron-Labs-Audex-30B-A3B (Audex), a unified audio-text LLM built on Nemotron-Cascade-2-30B-A3B, a strong text-only MoE LLM.
DuoTok is a source‑aware dual‑track music tokenizer designed for vocal‑accompaniment generation. It first learns a semantic audio representation via self‑supervised pretraining, then refines source‑aware structure with feature‑replacement noise and multi‑task supervision (spectral reconstruction, source separation regularization, and an ASR head for lyric alignment). The encoder is frozen and hard‑routed codebooks for vocals and accompaniment are learned, while a diffusion decoder restores fine acoustic detail from the discrete tokens, achieving a favorable predictability‑fidelity trade‑off at ultra‑low bitrate across public benchmarks.
By Rui Lin, Zhiyue Wu, Jiahe Lei, Kangdi Wang, Weixiong Chen, Junyu Dai, Tao Jiang
arXiv:2603. 08683v2 Announce Type: replace-cross Abstract: Autoregressive "language" models (LMs) trained on raw waveforms can be repurposed for lossless audio compression, but prior work is limited to 8-bit audio, leaving open whether such approaches work for practical settings (16/24-bit) and can compete with existing codecs.
By Phillip Long, Zachary Novack, Chris Donahue