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.

What it solves

MOSS-TTS Family provides a suite of production-ready speech and sound generation models designed to handle complex real-world audio needs. It addresses the limitation where a single TTS model cannot simultaneously provide high-fidelity long-form speech, expressive multi-speaker dialogues, zero-shot voice design from text, low-latency real-time interaction, and environmental sound effects.

How it works

The project breaks the audio generation workflow into five specialized models based on different architectures:

  • MOSS-TTS: A flagship model for high-fidelity, long-form speech and zero-shot voice cloning with fine-grained control over pronunciation and duration.
  • MOSS-TTSD: A specialized model for expressive, multi-speaker, ultra-long dialogues.
  • MOSS-VoiceGenerator: A voice design model that creates diverse voices and styles from text prompts without needing reference audio.
  • MOSS-TTS-Realtime: A low-latency model for voice agents that uses incremental synthesis and context-awareness to maintain coherence across turns.
  • MOSS-SoundEffect: A model for generating environmental and biological sound effects from text.

It utilizes complementary architectures like MossTTSDelay (for stability and speed) and MossTTSLocal (for lightweight streaming), and supports 48 kHz stereo output via the MOSS-Audio-Tokenizer-v2.

Who it’s for

This is for developers building voice agents, content creators producing films or games, and researchers working on speech synthesis and audio generation.

Highlights

  • Comprehensive Audio Suite: Covers everything from speech synthesis and dialogue to voice design and audio effects.
  • Multilingual Support: Supports 31 languages, including code-switching capabilities.
  • High Fidelity: Native 48 kHz stereo input and output.
  • Flexible Deployment: Supports PyTorch-free inference via llama.cpp and ONNX Runtime, as well as accelerated backends like SGLang-Omni and vLLM-Omni.
  • Real-time Performance: MOSS-TTS-Realtime achieves a Time To First Byte (TTFB) of 180 ms.

Related

  • Project
    OpenMOSS/MOSS-TTSD

    MOSS-TTSD is a long-form spoken dialogue generation model that converts dialogue scripts into expressive, multi-party conversational speech for podcasts and audiobooks.

  • Project
    OpenMOSS/MOSS-TTS-Nano

    A multilingual tiny speech generation model with 0.1B parameters designed for real-time, CPU-based inference and high-fidelity audio output.

  • Project
    2noise/ChatTTS

    ChatTTS is a generative text-to-speech model optimized for conversational dialogue, providing natural, expressive speech with control over laughter and pauses.

  • Project
    resemble-ai/chatterbox

    A family of open-source text-to-speech models providing low-latency voice cloning and multilingual speech generation across various model sizes for GPU and CPU deployment.

  • Project
    Audio8-AI/Audio8_TTS

    A compact 0.6B-parameter multilingual text-to-speech model featuring zero-shot voice cloning and efficient deployment options for CPU and GPU.