# GRAFT: Grafted Reference Audio for Fine-grained Pronunciation in Zero-shot Text-to-Speech

Published: 2026-07-07 · Updated: 2026-07-07 · Category: Research

Source page: https://www.sipply.dev/story/graft-grafted-reference-audio-for-fine-grained-pronunciation-8407f00d · 中文: https://www.sipply.dev/zh/story/graft-grafted-reference-audio-for-fine-grained-pronunciation-8407f00d

## Summary

Researchers have unveiled GRAFT, a novel mechanism designed to improve pronunciation accuracy in zero-shot text-to-speech (TTS) systems. By conditioning the TTS models on per-word audio samples, GRAFT significantly reduces phoneme error rates, demonstrating superior performance compared to existing zero-shot systems. This advancement not only enhances the clarity and naturalness of synthesized speech but also maintains the speaker's unique characteristics, making it a promising development in the field of speech synthesis.

## Key takeaways

- GRAFT improves pronunciation in TTS by using per-word audio samples.
- The new mechanism achieves lower phoneme error rates than previous systems.
- GRAFT maintains speaker similarity while enhancing speech accuracy.

## Why it matters

The development of GRAFT represents a significant step forward in the capabilities of text-to-speech technology, potentially transforming applications in accessibility, education, and entertainment by providing more accurate and natural-sounding speech synthesis.

- **Product managers:** For product managers, GRAFT's advancements in TTS technology could lead to more engaging user experiences and open new market opportunities in sectors that rely on high-quality speech synthesis.
- **Developers:** Developers can leverage GRAFT to create more sophisticated TTS applications that require precise pronunciation, enhancing the overall performance and user satisfaction of their products.
- **Students and job seekers:** Students and job-seekers in AI and machine learning can benefit from understanding GRAFT's innovative approach, as it highlights emerging trends in speech synthesis and offers insights into potential career opportunities in this evolving field.

## Sources

- [GRAFT: Grafted Reference Audio for Fine-grained Pronunciation in Zero-shot Text-to-Speech](https://arxiv.org/abs/2607.02633) (arXiv cs.LG, 2026-07-07)

## 中文

### GRAFT：用于零样本文本到语音的细粒度发音的嫁接参考音频

研究人员揭示了GRAFT，这是一种新机制，旨在提高零样本文本到语音（TTS）系统中的发音准确性。通过对每个单词的音频样本进行条件处理，GRAFT显著降低了音素错误率，表现出优于现有零样本系统的性能。这一进展不仅增强了合成语音的清晰度和自然性，还保持了说话者的独特特征，使其成为语音合成领域的一个有前景的发展。

**要点：**

- GRAFT通过使用每个单词的音频样本来改善TTS中的发音。
- 新机制实现了比以前系统更低的音素错误率。
- GRAFT在提高语音准确性的同时保持了说话者的相似性。

**为什么重要：** GRAFT的发展代表了文本到语音技术能力的重要进步，可能通过提供更准确和自然的语音合成来改变无障碍、教育和娱乐等应用领域。

---
Sipply: Sip AI News Simply. https://www.sipply.dev