AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Scaling Speech Tokenizers with Diffusion Autoencoders

Yuancheng Wang, Zhenyu Tang, Yun Wang, Arthur Hinsvark, Yingru Liu, Yinghao Aaron Li, Kainan Peng, Junyi Ao, Mingbo Ma, Mike Seltzer, Qing He, Xubo Liu

ICLR 2026 · Image generation › Autoregressive and latent generation · approach: Neural text-to-speech synthesis · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add CTC semantic regularization (weight 0.1) (ablation)SiTok (12.5 Hz, 1 codebook) · SeedTTS test-enWER (%): 33.0 -> 4.06Improved
add CTC semantic regularization (ablation)SiTok · LLM ASRWER: 29.4 -> 4.95Improved
raise CTC weight from 0.1 to 1 (ablation)SiTok · SeedTTS test-enWER (%): 4.06 -> 10.1Hurt
use diffusion loss instead of L1 regression loss (ablation)SiTok · SeedTTS test-enWER (%): 4.66 -> 4.06Improved
train diffusion end-to-end rather than finetuning only the decoder on a regression model (ablation)SiTok · SeedTTS test-enWER (%): 5.73 -> 4.06Improved
add token classifier-free guidance (ablation)SiTok (CN=1) · SeedTTS test-enWER (%): 4.06 -> 3.34Improved
add decoder finetuning (ablation)SiTok (CN=1) · SeedTTS test-enspeaker similarity: 0.641 -> 0.682Improved
increase codebook size from 2^13 to 2^17 (ablation)SiTok · SeedTTS test-enWER (%): 5.48 -> 3.94Improved
increase number of RVQ codebooks from 1 to 8 (ablation)SiTok · SeedTTS test-enWER (%): 4.30 -> 2.50Improved
increase frame rate from 6.25 Hz to 12.5 Hz (ablation)SiTok · SeedTTS test-enWER (%): 23.0 -> 4.06Improved
scale model from L (1.12B) to XL (1.61B) (ablation)SiTok · Understanding tasksSV EER: 13.8 -> 14.7Hurt
scale model from S (0.63B) to XL (1.61B) (ablation)SiTok · SeedTTS test-enWER (%): 4.18 -> 3.84Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy