# Same Bit Width, Different Outcomes: Post-Training Quantization of Text-to-Speech Across Architectures

arXiv: https://arxiv.org/abs/2609.28974 · HTML: https://mlsanity.com/paper/2609.28974/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 32 references that are in the map.
- First posted on arXiv 24 Sep 2026.

## Builds on (most cited first, 23 of 23)

- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models](https://mlsanity.com/paper/2211.10438/) (2022)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://mlsanity.com/paper/2306.00978/) (2023)
- [Scaling Speech Technology to 1,000+ Languages](https://mlsanity.com/paper/2305.13516/) (2023)
- [HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precision](https://mlsanity.com/paper/1905.03696/) (2019)
- [UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022](https://mlsanity.com/paper/2204.02152/) (2022)
- [F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching](https://mlsanity.com/paper/2410.06885/) (2024)
- [Q-Diffusion: Quantizing Diffusion Models](https://mlsanity.com/paper/2302.04304/) (2023)
- [Post-training Quantization on Diffusion Models](https://mlsanity.com/paper/2211.15736/) (2022)
- [DNSMOS P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors](https://mlsanity.com/paper/2110.01763/) (2021)
- [StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models](https://mlsanity.com/paper/2306.07691/) (2023)
- [VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning](https://mlsanity.com/paper/2509.24650/) (2025)
- [Deep Learning Based Assessment of Synthetic Speech Naturalness](https://mlsanity.com/paper/2104.11673/) (2021)
- [LightSpeech: Lightweight and Fast Text to Speech with Neural Architecture Search](https://mlsanity.com/paper/2102.04040/) (2021)
- [Streaming Sequence-to-Sequence Learning with Delayed Streams Modeling](https://mlsanity.com/paper/2509.08753/) (2025)
- [OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models](https://mlsanity.com/paper/2604.00688/) (2026)
- [MPQ-DM: Mixed Precision Quantization for Extremely Low Bit Diffusion Models](https://mlsanity.com/paper/2412.11549/) (2024)
- [On-device neural speech synthesis](https://mlsanity.com/paper/2109.08710/) (2021)
- [PTQ4ADM: Post-Training Quantization for Efficient Text Conditional Audio Diffusion Models](https://mlsanity.com/paper/2409.13894/) (2024)
- [Post-Training Quantization for Audio Diffusion Transformers](https://mlsanity.com/paper/2510.00313/) (2025)
- [Compact Neural TTS Voices for Accessibility](https://mlsanity.com/paper/2501.17332/) (2025)
- [BitTTS: Highly Compact Text-to-Speech Using 1.58-bit Quantization and Weight Indexing](https://mlsanity.com/paper/2506.03515/) (2025)

## Abstract (arXiv)

Post-training quantization (PTQ) reduces the cost of on-device text-to-speech (TTS), but published evaluations cover one system or method. We evaluate PTQ across TTS architectures under one protocol with three core models, weight and activation ablations of eight more, and two held-out models quantized blind. Four-bit per-channel weights reduce UTMOS, a predicted mean opinion score, by 2.8 on Supertonic and 0.07 on Kokoro, and per-tensor scaling can cause severe degradation even at 8 bits. The same bit width yields different outcomes, because the sensitive component is model-specific and not reliably predicted from the model class. A staged ablation procedure identifies it, and per-layer GPTQ can restore it to within 0.1 UTMOS. Real int8 and int4 kernels reproduce the simulated ordering at hardware-dependent cost. On a Mac mini, a 4-bit weight kernel runs Supertonic at 0.60x the fp32 latency while int8 is slower, so each configuration requires validation on the target runtime.