DiTReducio: A Training-Free Acceleration for DiT-Based TTS via Progressive Calibration
On the map: Diffusion models → Audio, speech and music
Abstract (arXiv)
While Diffusion Transformers (DiT) have advanced non-autoregressive (NAR) speech synthesis, their high computational demands remain an limitation. Existing DiT-based text-to-speech (TTS) model acceleration approaches mainly focus on reducing sampling steps through distillation techniques, yet they remain constrained by training costs. We introduce DiTReducio, a training-free acceleration framework that compresses computations in DiT-based TTS models via progressive calibration. We propose two compression methods, Temporal Skipping and Branch Skipping, to eliminate redundant computations during inference. Moreover, based on two characteristic attention patterns identified within DiT layers, we devise a pattern-guided strategy to selectively apply the compression methods. Our method allows flexible modulation between generation quality and computational efficiency through adjustable compression thresholds. Experimental evaluations conducted on F5-TTS and MegaTTS 3 demonstrate that DiTReducio achieves a 75.4% reduction in FLOPs and improves the Real-Time Factor (RTF) by 37.1%, while preserving generation quality.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Audio, speech and music, one of 14 directions in Diffusion models. Builds on 39 of its 56 references that are in the map. First posted on arXiv 11 Sep 2025.
Builds on
39 of 56 references.
Closest ancestors · 11
Same direction, newest first
- 2025ZipVoice: Fast and High-Quality Zero-Shot Text-to-Speech with Flow Matching16 Jun
- 2025CosyVoice 3: Towards In-the-wild Speech Generation via Scaling-up and Post-training23 May
- 2025OZSpeech: One-step Zero-shot Speech Synthesis with Learned-Prior-Conditioned Flow Matching19 May
- 2025MegaTTS 3: Sparse Alignment Enhanced Latent Diffusion Transformer for Zero-Shot Speech Synthesis26 Feb
- 2024F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching9 Oct
- 2023ReFlow-TTS: A Rectified Flow Model for High-fidelity Text-to-Speech29 Sep
- 2023Matcha-TTS: A fast TTS architecture with conditional flow matching6 Sep
- 2022ProDiff: Progressive Fast Diffusion Model For High-Quality Text-to-Speech13 Jul
- 2022DiffGAN-TTS: High-Fidelity and Efficient Text-to-Speech with Denoising Diffusion GANs28 Jan
- 2021Guided-TTS: A Diffusion Model for Text-to-Speech via Classifier Guidance23 Nov
- 2021Diff-TTS: A Denoising Diffusion Model for Text-to-Speech3 Apr
Other references · 28
- 2022Robust Speech Recognition via Large-Scale Weak Supervisioncited by 4,057
- 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
- 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Stepscited by 1,979
- 2020Big Bird: Transformers for Longer Sequencescited by 1,684
- 2022Token Merging: Your ViT But Fastercited by 1,021
- 2023DeepCache: Accelerating Diffusion Models for Freecited by 359
- 2024CosyVoice: A Scalable Multilingual Zero-shot Text-to-speech Synthesizer based on Supervised Semantic Tokenscited by 334
- 2024Seed-TTS: A Family of High-Quality Versatile Speech Generation Modelscited by 300
- 2024PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reductioncited by 294
- 2022Post-training Quantization on Diffusion Modelscited by 267
- 2023Token Merging for Fast Stable Diffusioncited by 224
- 2023PTQD: Accurate Post-Training Quantization for Diffusion Modelscited by 170
16 more
- 2024MaskGCT: Zero-Shot Text-to-Speech with Masked Generative Codec Transformercited by 161
- 2024E2 TTS: Embarrassingly Easy Fully Non-Autoregressive Zero-Shot TTScited by 152
- 2024Real-Time Video Generation with Pyramid Attention Broadcastcited by 149
- 2024Learning-to-Cache: Accelerating Diffusion Transformer via Layer Cachingcited by 142
- 2024VALL-E 2: Neural Codec Language Models are Human Parity Zero-Shot Text to Speech Synthesizerscited by 122
- 2024DiTFastAttn: Attention Compression for Diffusion Transformer Modelscited by 115
- 2024Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Modelscited by 114
- 2021Large-scale Self-Supervised Speech Representation Learning for Automatic Speaker Verificationcited by 78
- 2024LD-Pruner: Efficient Pruning of Latent Diffusion Models using Task-Agnostic Insightscited by 65
- 2024ELLA-V: Stable Neural Codec Language Modeling with Alignment-guided Sequence Reorderingcited by 50
- 2023LibriSpeech-PC: Benchmark for Evaluation of Punctuation and Capitalization Capabilities of end-to-end ASR Modelscited by 49
- 2023Make-A-Voice: Unified Voice Synthesis With Discrete Representationcited by 27
- 2024VoxInstruct: Expressive Human Instruction-to-Speech Generation with Unified Multilingual Codec Language Modellingcited by 21
- 2025DiTFastAttnV2: Head-wise Attention Compression for Multi-Modality Diffusion Transformerscited by 20
- 2025Cached Adaptive Token Merging: Dynamic Token Reduction and Redundant Computation Elimination in Diffusion Modelcited by 16
- 2025Pruning for Sparse Diffusion Models based on Gradient Flowcited by 5
Built on by
Citing papers per year: 2025: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2024DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors
- 2026DiTAR+: Dual Optimization for Robust Autoregressive Diffusion Speech Synthesis
- 2025Accelerating Diffusion Transformer-Based Text-to-Speech with Transformer Layer Caching
- 2025M3-TTS: Multi-modal DiT Alignment & Mel-latent for Zero-shot High-fidelity Speech Synthesis
- 2026DLLM-TTS: Block Discrete Diffusion Language Model for Text-to-Speech Synthesis
- 2025DiTAR: Diffusion Transformer Autoregressive Modeling for Speech Generation
- 2025Accelerating Flow-Matching-Based Text-to-Speech via Empirically Pruned Step Sampling
- 2024SimpleSpeech 2: Towards Simple and Efficient Text-to-Speech with Flow-based Scalar Latent Transformer Diffusion Models