Jing-Bin Hu

Timbre drift in multi-turn expressive voice agents can be eliminated by replacing implicit context embeddings with explicitly supervised style instructions optimized via rejection sampling and context-aware DPO.

StreamTN is proposed, a lightweight LLM-based Chinese streaming TN model that employs a dual-track streaming framework in which input tokens and output tokens are processed on two parallel tracks, enabling low-latency real-time inference without complex prompting.

A preference optimization framework with Large Audio-Language Model (LALM) feedback for controllable non-verbal vocalization (NVV) generation in continuous autoregressive speech models that enables DPO-style preference learning without explicit sequence likelihoods while preserving direct supervision on preferred realizations is proposed.

FlashTTS slashes First-Packet Latency to 325ms, revolutionizing real-time speech dialogue systems without sacrificing voice quality.

MeanVC 2 cuts voice conversion latency in half while enhancing robustness to low-quality audio references, revolutionizing real-time voice applications.

Open-source TTS models can beat commercial systems in specific languages, but current instruction-following TTS still struggles with complex instructions like nuanced paralinguistic controls.

Achieve faster, more accurate turn-taking in spoken dialogue by fusing streaming speech recognition with raw audio cues – even when it's noisy.