DLLM-TTS framework enables faster, more accurate text-to-speech synthesis

Researchers have introduced DLLM-TTS, a novel text-to-speech synthesis framework that addresses the trade-off between speech intelligibility and generation speed. This new approach formulates TTS as conditional block discrete diffusion over X-Codec2 neural audio codec tokens. By decomposing sequences into blocks and applying masked diffusion within each block, DLLM-TTS achieves efficient, parallel generation with a real-time factor of 0.15, while maintaining competitive performance on benchmarks like Seed-TTS-eval. AI

IMPACT This framework offers a potential path to more efficient and data-efficient speech synthesis, improving real-time generation capabilities.

RANK_REASON The cluster describes a new research paper detailing a novel framework for text-to-speech synthesis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• DLLM-TTS
• Hugging Face
• Nityanand Mathur
• Seed-TTS Eval
• X-Codec2

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →