{
  "hf_id": "OpenMOSS-Team/MOSS-TTS",
  "meta": {
    "title": "MOSS-TTS",
    "slug": "moss-tts",
    "provider": "OpenMOSS",
    "description": "Flagship 8B model of the OpenMOSS MOSS-TTS Family — high-fidelity zero-shot voice cloning, ultra-long stable speech, token-level duration and phoneme control, 20-language code-switched synthesis — served via vLLM-Omni through the OpenAI /v1/audio/speech API (24 kHz mono).",
    "date_added": "2026-06-17",
    "date_updated": "2026-06-15",
    "tasks": [
      "omni"
    ],
    "related_recipes": [
      "OpenMOSS-Team/MOSS-TTS-Realtime",
      "OpenMOSS-Team/MOSS-TTSD-v1.0",
      "OpenMOSS-Team/MOSS-VoiceGenerator"
    ]
  },
  "model": {
    "model_id": "OpenMOSS-Team/MOSS-TTS",
    "min_vllm_version": "0.22.0",
    "architecture": "dense",
    "parameter_count": "8B",
    "active_parameters": "8B",
    "context_length": 0,
    "base_args": [
      "--omni",
      "--deploy-config",
      "vllm_omni/deploy/moss_tts.yaml"
    ],
    "base_env": {},
    "install": {
      "pip": {
        "command": "uv venv\nsource .venv/bin/activate\nuv pip install -U vllm --torch-backend auto"
      },
      "docker": {
        "command": "docker pull vllm/vllm-openai:latest"
      },
      "extras": [
        {
          "note": "vLLM-Omni provides the TTS serving backend and the `--omni` serve path. See the installation guide for pre-built wheels or building from source.",
          "command": "uv pip install git+https://github.com/vllm-project/vllm-omni.git"
        }
      ]
    }
  },
  "omni": {
    "tasks": [
      {
        "id": "t2a",
        "label": "Text → Speech",
        "endpoint": "/v1/audio/speech",
        "description": "MOSS-TTS zero-shot voice cloning via /v1/audio/speech (24 kHz mono)",
        "curl": "curl -X POST http://localhost:8000/v1/audio/speech \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"model\": \"OpenMOSS-Team/MOSS-TTS\",\n    \"input\": \"Hello, this is a voice cloning test.\",\n    \"voice\": \"default\",\n    \"ref_audio\": \"https://raw.githubusercontent.com/OpenMOSS/MOSS-TTS/main/assets/audio/zh_1.wav\",\n    \"response_format\": \"wav\"\n  }' --output output.wav\n"
      }
    ]
  },
  "features": {},
  "opt_in_features": [],
  "variants": {
    "default": {
      "precision": "bf16",
      "vram_minimum_gb": 32,
      "description": "MOSS-TTS 8B (MossTTSDelay) — flagship production TTS (reference run on 1x H100 80 GB)"
    }
  },
  "hardware_overrides": {},
  "guide": "## Overview\n\n[MOSS-TTS](https://huggingface.co/OpenMOSS-Team/MOSS-TTS) is the **flagship 8B\nmodel** of OpenMOSS's [MOSS-TTS Family](https://huggingface.co/OpenMOSS-Team/MOSS-TTS),\na production-grade TTS foundation model served through **vLLM-Omni** with the\nOpenAI-compatible `/v1/audio/speech` API. It delivers high-fidelity zero-shot\nvoice cloning, **ultra-long stable speech (up to ~1 hour)**, token-level\nduration control, Pinyin/IPA phoneme control, and **20-language** /\ncode-switched synthesis. Output is **24 kHz mono** via the shared\n`OpenMOSS-Team/MOSS-Audio-Tokenizer` codec (~7 GB, auto-downloaded; override\nwith `MOSS_TTS_CODEC_PATH`).\n\nThe MOSS-TTS Family ships five production-ready models, each with its own\nrecipe: **MOSS-TTS** (this page, flagship TTS), **MOSS-TTS-Realtime**\n(low-latency streaming), **MOSS-TTSD** (multi-speaker dialogue),\n**MOSS-VoiceGenerator** (zero-shot voice design), and **MOSS-SoundEffect**\n(sound-effect synthesis).\n\nMOSS-TTS is released in two architectures: the 8B `MossTTSDelay` (this\ncheckpoint, production-recommended) and a 1.7B `MossTTSLocal`\n(`OpenMOSS-Team/MOSS-TTS-Local-Transformer`) tuned for streaming-oriented\nsystems. A `MOSS-TTS-v1.5` point release adds 31-language support and steadier\ncloning with `[pause Xs]` markers (same serving path; set `language` for best\nresults).\n\n## Prerequisites\n\n- **Hardware**: a single CUDA GPU. Reference run: 1x H100 80 GB — ~18 GB for\n  the 8B talker + ~8 GB for the codec decoder on the same device\n  (`gpu_memory_utilization: 0.85` in `moss_tts.yaml`).\n- **vLLM-Omni** targeting vLLM >= 0.22.\n\n## Installation\n\n```bash\nuv venv && source .venv/bin/activate\nuv pip install git+https://github.com/vllm-project/vllm-omni.git\n```\n\n## Launch the server\n\nAll MossTTSDelay checkpoints (MOSS-TTS, MOSS-TTSD, MOSS-SoundEffect,\nMOSS-VoiceGenerator) share `model_type=moss_tts_delay`, so the deploy config\nmust be passed explicitly to select the right pipeline:\n\n```bash\nvllm serve OpenMOSS-Team/MOSS-TTS --omni \\\n  --deploy-config vllm_omni/deploy/moss_tts.yaml --port 8000\n```\n\n## Client usage\n\n### Voice cloning (curl)\n\n```bash\ncurl -X POST http://localhost:8000/v1/audio/speech \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"model\": \"OpenMOSS-Team/MOSS-TTS\",\n    \"input\": \"Hello, this is a voice cloning test.\",\n    \"voice\": \"default\",\n    \"ref_audio\": \"https://raw.githubusercontent.com/OpenMOSS/MOSS-TTS/main/assets/audio/zh_1.wav\",\n    \"response_format\": \"wav\"\n  }' --output output.wav\n```\n\n## Known limitations\n\n- Output is **24 kHz mono**.\n- `MOSS_TTS_CODEC_PATH` overrides the codec checkpoint location if you have a\n  local copy.\n\n## References\n\n- [MOSS-TTS model card](https://huggingface.co/OpenMOSS-Team/MOSS-TTS)\n- [MOSS-TTS Family](https://huggingface.co/OpenMOSS-Team/MOSS-TTS) · [GitHub](https://github.com/OpenMOSS/MOSS-TTS)\n- [vLLM-Omni MOSS-TTS offline example](https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/text_to_speech/moss_tts)\n",
  "source": {
    "source": "vllm-omni",
    "repo": "vllm-project/vllm-omni",
    "ref": "main",
    "sha": "beeedd99df33b432b7cf74caa8b28e1893124827",
    "path": "recipes/models/OpenMOSS-Team/MOSS-TTS.yaml",
    "edit_url": "https://github.com/vllm-project/vllm-omni/edit/main/recipes/models/OpenMOSS-Team/MOSS-TTS.yaml",
    "overlays": []
  }
}