MOSS-TTS and MOSS-TTSD
MOSS-TTSD is a specialized extension of MOSS-TTS that adds dialogue-specific capabilities (long-context modeling, multi-speaker synthesis) on top of the core TTS functionality, making them ecosystem siblings where MOSS-TTSD builds upon MOSS-TTS for conversational applications.
About MOSS-TTS
OpenMOSS/MOSS-TTS
MOSS‑TTS Family is an open‑source speech and sound generation model family from MOSI.AI and the OpenMOSS team. It is designed for high‑fidelity, high‑expressiveness, and complex real‑world scenarios, covering stable long‑form speech, multi‑speaker dialogue, voice/character design, environmental sound effects, and real‑time streaming TTS.
# Technical Summary Built on a modular architecture, MOSS-TTS decomposes speech synthesis into five specialized models—flagship TTS for zero-shot voice cloning with phoneme-level control, a dialogue model outperforming closed-source baselines on objective metrics, a prompt-based voice generator requiring no reference audio, a low-latency realtime agent model (180ms TTFB), and a sound effect generator. The framework supports multiple inference backends including PyTorch-free deployment via llama.cpp with GGUF quantization and ONNX audio codec decoding, plus SGLang acceleration achieving 3× faster generation throughput. Models are available on Hugging Face and ModelScope with fine-tuning tutorials and REST API documentation via the MOSI.AI studio platform.
About MOSS-TTSD
OpenMOSS/MOSS-TTSD
MOSS-TTSD is a spoken dialogue generation model designed for expressive multi-speaker synthesis. It features long-context modeling, flexible speaker control, and multilingual support, while enabling zero-shot voice cloning from short audio references.
Built on transformer-based architecture with audio tokenization via XY-Tokenizer, the model uses a continuation-based workflow where speaker reference audio and dialogue scripts enable seamless multi-speaker synthesis over extended contexts. Optimized for SGLang inference engine acceleration (up to 16x speedup), it supports streaming generation, fine-tuning via LoRA and full-parameter training, and integrates with Hugging Face model hub and spaces for easy deployment across 20 languages.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work