Confucius4-TTS
A multilingual, zero-shot text-to-speech system that clones a voice from a short reference clip and speaks any of fourteen supported languages in that voice, without needing a matching transcript.
Confucius4-TTS is a text-to-speech system built on a speech encoder combined with a large language model, designed for multilingual and cross-lingual speech synthesis. It clones a speaker's voice from a reference audio clip without requiring a matching transcript, and can carry that cloned voice and its emotional character across fourteen supported languages, including Chinese, English, Japanese, Korean, and several European and Southeast Asian languages.
It supports zero-shot voice transfer without additional training and aims for stable performance across real-world multilingual scenarios. Inference runs through Hugging Face Transformers by default, with an optional vLLM backend that accelerates the autoregressive text-to-semantic stage using paged attention for higher throughput.
It is written in Python and requires CUDA and a recent PyTorch and Transformers installation to run.