mlsanityData to 7 Oct 2026

Decoding Order Matters in Autoregressive Speech Synthesis

arXiv 2601.08450 ↗First posted 13 Jan 2026cs.SD cs.AI eess.AScited by 0

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

Autoregressive speech synthesis often adopts a left-to-right order, yet generation order is a modelling choice. We investigate decoding order through masked diffusion framework, which progressively unmasks positions and allows arbitrary decoding orders during training and inference. By interpolating between identity and random permutations, we show that randomness in decoding order affects speech quality. We further compare fixed strategies, such as textttl2r and textttr2l with adaptive ones, such as Top-K, finding that fixed-order decoding, including the dominating left-to-right approach, is suboptimal, while adaptive decoding yields better performance. Finally, since masked diffusion requires discrete inputs, we quantise acoustic representations and find that even 1-bit quantisation can support reasonably high-quality speech.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 23 of its 26 references that are in the map. First posted on arXiv 13 Jan 2026.

Builds on

23 of 26 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models12 Mar
  2. 2025Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions10 Feb
  3. 2024Simple and Effective Masked Diffusion Language Models11 Jun
  4. 2021Autoregressive Diffusion Models5 Oct

Other references · 19

  1. 2017Neural Discrete Representation Learningcited by 4,650
  2. 2016WaveNet: A Generative Model for Raw Audiocited by 1,710
  3. 2020HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesiscited by 1,139
  4. 2017Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictionscited by 800
  5. 2021SoundStream: An End-to-End Neural Audio Codeccited by 756
  6. 2023Neural Codec Language Models are Zero-Shot Text to Speech Synthesizerscited by 737
  7. 2022High Fidelity Neural Audio Compressioncited by 673
  8. 2024Autoregressive Image Generation without Vector Quantizationcited by 644
  9. 2021Grad-TTS: A Diffusion Probabilistic Model for Text-to-Speechcited by 445
  10. 2023Finite Scalar Quantization: VQ-VAE Made Simplecited by 425
  11. 2024MaskGCT: Zero-Shot Text-to-Speech with Masked Generative Codec Transformercited by 159
  12. 2024Fluid: Scaling Autoregressive Text-to-image Generative Models with Continuous Tokenscited by 147
7 more
  1. 2023Matcha-TTS: A fast TTS architecture with conditional flow matchingcited by 142
  2. 2024The T05 System for The VoiceMOS Challenge 2024: Transfer Learning from Deep Image Classifier to Naturalness MOS Prediction of High-Quality Synthetic Speechcited by 78
  3. 2025DiTAR: Diffusion Transformer Autoregressive Modeling for Speech Generationcited by 56
  4. 2025Recent Advances in Discrete Speech Tokens: A Reviewcited by 42
  5. 2024Autoregressive Diffusion Transformer for Text-to-Speech Synthesiscited by 34
  6. 2023DiffAR: Denoising Diffusion Autoregressive Model for Raw Speech Waveform Generationcited by 11
  7. 2025Optimal word order for non-causal text generation with Large Language Models: the Spanish casecited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Masked Diffusion Models are Secretly Learned-Order Autoregressive Models
  2. 2026Rethinking the Generation Order of Block Diffusion Language Models
  3. 2026DiTAR+: Dual Optimization for Robust Autoregressive Diffusion Speech Synthesis
  4. 2025Accelerating Autoregressive Speech Synthesis Inference With Speech Speculative Decoding
  5. 2026DELTA-TTS: Adapting Autoregressive Model into Diffusion Language Model for Text-to-Speech
  6. 2026Confidence-Based Decoding is Provably Efficient for Diffusion Language Models
  7. 2026Low-Confidence Remasking Traps Flexibility: Realizing Arbitrary-Order Potential for Diverse Rollouts in Diffusion LLMs
  8. 2026Generation Order and Parallel Decoding in Masked Diffusion Models: An Information-Theoretic Perspective