![[2602.19574] CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19574] CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment
Summary
The paper presents CTC-TTS, a novel dual-streaming text-to-speech system that utilizes a CTC-based aligner for improved text-speech alignment and lower latency, outperforming traditional methods.
Why It Matters
As demand for real-time speech synthesis grows, advancements in TTS technology are crucial. CTC-TTS offers a significant improvement in generating natural speech with low latency, making it relevant for applications in AI communication, virtual assistants, and accessibility tools.
Key Takeaways
- CTC-TTS replaces traditional GMM-HMM aligners with a CTC-based approach for better performance.
- The system introduces bi-word interleaving to enhance text-speech alignment.
- Two variants of CTC-TTS are designed for quality and latency optimization.
- Experiments indicate CTC-TTS outperforms fixed-ratio interleaving and MFA-based methods.
- Speech samples demonstrate the practical application of the proposed system.
Electrical Engineering and Systems Science > Audio and Speech Processing arXiv:2602.19574 (eess) [Submitted on 23 Feb 2026] Title:CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment Authors:Hanwen Liu, Saierdaer Yusuyin, Hao Huang, Zhijian Ou View a PDF of the paper titled CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment, by Hanwen Liu and 3 other authors View PDF HTML (experimental) Abstract:Large-language-model (LLM)-based text-to-speech (TTS) systems can generate natural speech, but most are not designed for low-latency dual-streaming synthesis. High-quality dual-streaming TTS depends on accurate text--speech alignment and well-designed training sequences that balance synthesis quality and latency. Prior work often relies on GMM-HMM based forced-alignment toolkits (e.g., MFA), which are pipeline-heavy and less flexible than neural aligners; fixed-ratio interleaving of text and speech tokens struggles to capture text--speech alignment regularities. We propose CTC-TTS, which replaces MFA with a CTC based aligner and introduces a bi-word based interleaving strategy. Two variants are designed: CTC-TTS-L (token concatenation along the sequence length) for higher quality and CTC-TTS-F (embedding stacking along the feature dimension) for lower latency. Experiments show that CTC-TTS outperforms fixed-ratio interleaving and MFA-based baselines on streaming synthesis and zero-shot tasks. Speech samples are available at this https URL. Comments: ...