OpenMOSS/MOSS-TTSD

A multilingual model for long-form, multi-speaker dialogue synthesis with flexible speaker control and zero-shot voice cloning

What it solves

MOSS-TTSD addresses the limitation of traditional text-to-speech (TTS) systems that focus on single-speaker monologues. It transforms static dialogue scripts into continuous, expressive multi-party conversations, maintaining speaker identity and emotional flow over long durations, which is essential for professional content like podcasts and audiobooks.

How it works

The model uses a continuation workflow where users provide reference audio and transcripts for each speaker (up to 5 speakers). It then generates audio based on a dialogue script, ensuring natural turn-taking and overlapping speech patterns. For high-performance deployment, it can be fused with the MOSS-Audio-Tokenizer and run via the SGLang inference engine to accelerate generation.

Who it’s for

It is designed for creators and developers producing long-form spoken content, such as AI podcasts, sports and esports commentary, dubbing, and audiobook production.

Highlights

  • Multi-party Dialogue: Supports 1 to 5 speakers with flexible control over conversational rhythms and persona maintenance.
  • Extreme Long-Context: Capable of generating up to 60 minutes of coherent audio in a single session while maintaining consistent identity.
  • Zero-Shot Voice Cloning: High-fidelity voice cloning using only short reference audio samples.
  • Multilingual Support: Robust performance across 20 languages, including Chinese, English, Japanese, and various European languages.
  • High-Speed Inference: Integration with SGLang allows for significantly accelerated end-to-end inference.

Related

  • Project
    OpenMOSS/MOSS-TTS

    A family of open-source speech and sound generation models providing high-fidelity TTS, multi-speaker dialogue, voice design, real-time streaming, and sound effect generation.

  • 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
    fishaudio/fish-speech

    Fish Speech is an open‑source, large‑model‑powered text‑to‑speech system offering expressive, multilingual voice cloning, multi‑speaker scene generation, and on‑the‑fly audio editing. It provides two models – Drama 3 (35 B) for long‑form, high‑fidelity drama production, and S2‑Pro (4 B) for faster, consumer‑GPU deployment – with command‑line, API, and Web UI interfaces, Docker support, and detailed installation guides.

  • Project
    FireRedTeam/FireRedTTS3

    A unified speech generation and editing system that supports zero-shot voice cloning across 24 languages and 21 Chinese dialects, as well as instruction-based voice design and audio editing.

  • Project
    RVC-Boss/GPT-SoVITS

    A powerful few-shot voice conversion and text-to-speech system that can clone a voice using as little as 5 seconds of audio samples.