# TTS-Q: Speaker-Aware Post-Training Quantization of Text-to-Speech Diffusion Transformers

DevFeed: [TTS-Q: Speaker-Aware Post-Training Quantization of Text-to-Speech Diffusion Transformers](<https://devfeed.tech/articles/tts-q-speaker-aware-post-training-quantization-of-text-to-speech-diffusion-transformers-94183.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.12474>)

Author: Xiaoqin Feng, Yuqi Li, Weilun Feng, Yi-Cheng Lin, Kai Li, Xianglong Wang, Kuo Yang, Chuanguang Yang, Yingli Tian, Wei Li, Zhongwei Cheng, Lin Chen

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [quantization](<https://devfeed.tech/topics/quantization.md>), [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>), [voice ai](<https://devfeed.tech/topics/voice-ai.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-sd](<https://devfeed.tech/tags/cs-sd.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-transformers](<https://devfeed.tech/tags/diffusion-transformers.md>), [eess-as](<https://devfeed.tech/tags/eess-as.md>), [post-training](<https://devfeed.tech/tags/post-training.md>), [quantization](<https://devfeed.tech/tags/quantization.md>), [speaker](<https://devfeed.tech/tags/speaker.md>), [speech](<https://devfeed.tech/tags/speech.md>), [text-to-speech](<https://devfeed.tech/tags/text-to-speech.md>), [tts](<https://devfeed.tech/tags/tts.md>)

## AI overview

The abstract reports that 4-bit weight-and-activation quantization preserves TTS intelligibility fairly well but degrades speaker timbre. In tests on 1B- and 3.5B-parameter diffusion transformers, both weight and activation quantization contribute to the timbre loss; only activation errors show timestep structure that a precision schedule can target. TTS-Q selects difficult items from task-matched calibration candidates and assigns higher precision to late-step feed-forward activations within a balanced precision budget. The authors report the strongest timbre recovery at both scales without reducing intelligibility. A separately calibrated W4A4 reference reduces peak GPU memory; the excerpt gives no numerical memory or quality results.

## Source excerpt

arXiv:2610.12474v3 Announce Type: replace-cross Abstract: Zero-shot text-to-speech (TTS) can clone a voice from seconds of audio. Yet the quality comes from billion-parameter diffusion transformers whose size and multi-step sampling put them out of reach of the on-device, assistive, and real-time applications that need them most. Post-training quantization (PTQ) shrinks such models without retraining, but it has been built for language and vision, not speech. At 4 bits, a strong quantizer leaves intelligibility close to full precision but consistently erodes speaker timbre, so error rates stay low while the voice drifts from the target speaker. We ask why quantization hurts speech, where the damage lives, and whether it can be repaired without retraining. To find out, we run the first 4-bit weight-and-activation (W4A4) PTQ study of a TTS diffusion transformer at 1B and 3.5B. The timbre cost comes from both the weight and the activation quantization, but only the activation part has the timestep structure a precision schedule can exploit. We therefore introduce TTS-Q, which selects hard-to-quantize items from task-matched calibration candidates and applies a budget-balanced mixed-precision schedule that assigns higher precision to late-step feed-forward activations. TTS-Q recovers timbre best at both scales, at no cost to intelligibility. Separately, the deployed calibrated-W4A4 reference reduces peak memory on GPU, providing a practical reference for resource-constrained TTS deployment.