**[Xiaohongshu Open-Sources Fully Continuous Autoregressive TTS Model dots.tts, Supporting Zero-Shot Voice Cloning]** According to monitoring by Beating, Xiaohongshu's hi lab has open-sourced a 2-billion-parameter end-to-end autoregressive text-to-speech (TTS) model, dots.tts, under the Apache 2.0 license, including the complete inference and fine-tuning code. The released weights include the base pre-trained version, the self-correcting alignment (SCA) fine-tuned version, and the low-latency inference distilled version. Unlike traditional TTS architectures that rely on discrete audio codec tokens (e.g., VALL-E, CosyVoice, ChatTTS), dots.tts achieves a fully continuous, end-to-end autoregressive flow-matching architecture, completely avoiding the use of any discrete tokens throughout the pipeline. dots.tts combines continuous features extracted by a 48 kHz AudioVAE with a semantic encoder, a backbone language model (initialized from Qwen2.5-1.5B-Base, directly processing BPE text without requiring pinyin input), and an autoregressive flow-matching acoustic head to predict continuous latent variables, which are then reconstructed into audio by the generator. By directly predicting continuous features, dots.tts avoids the audio quality degradation caused by discrete quantization, preserving pronunciation details, timbre similarity, and emotional expressiveness. dots.tts is pre-trained on approximately 1.5 million hours of speech data. In the Seed-TTS-Eval evaluation, dots.tts achieved word error rates (WER) of 0.94% / 1.30% / 6.60% on Chinese, English, and Chinese challenging test sets, respectively, with similarity scores (SIM) of 81.0 / 77.1 / 79.5, reaching open-source state-of-the-art (SOTA) levels. In the MiniMax Multilingual benchmark across 24 languages, the average speaker similarity score reached 83.9. Xiaohongshu has provided a Gradio demo space on Hugging Face, allowing users to test zero-shot voice cloning online. [Original Link]