{
  "hf_id": "OpenMOSS-Team/MOSS-TTS-Realtime",
  "meta": {
    "title": "MOSS-TTS-Realtime",
    "slug": "moss-tts-realtime",
    "provider": "OpenMOSS",
    "description": "OpenMOSS's 1.7B real-time streaming TTS for low-latency voice agents (TTFB ~180 ms) — multi-turn context-aware incremental 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",
      "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
      "openbmb/VoxCPM2"
    ]
  },
  "model": {
    "model_id": "OpenMOSS-Team/MOSS-TTS-Realtime",
    "min_vllm_version": "0.22.0",
    "architecture": "dense",
    "parameter_count": "1.7B",
    "active_parameters": "1.7B",
    "context_length": 0,
    "base_args": [
      "--omni"
    ],
    "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-Realtime low-latency streaming TTS via /v1/audio/speech",
        "curl": "curl -X POST http://localhost:8000/v1/audio/speech \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"model\": \"OpenMOSS-Team/MOSS-TTS-Realtime\",\n    \"input\": \"This is a low-latency streaming TTS 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    \"stream\": true\n  }' --output output.wav\n"
      }
    ]
  },
  "features": {},
  "opt_in_features": [],
  "variants": {
    "default": {
      "precision": "bf16",
      "vram_minimum_gb": 16,
      "description": "MOSS-TTS-Realtime 1.7B (MossTTSRealtime) — lowest-latency streaming variant (reference run on 1x A10G 24 GB)"
    }
  },
  "hardware_overrides": {},
  "guide": "## Overview\n\n[MOSS-TTS-Realtime](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Realtime) is\nthe **1.7B real-time streaming** member of OpenMOSS's MOSS-TTS Family, served\nthrough **vLLM-Omni** with the OpenAI-compatible `/v1/audio/speech` API. It is\na multi-turn, context-aware model for **low-latency voice agents**, using a\nlocal depth transformer (no delay warm-up) for incremental synthesis. TTFB\n(time to first byte) is **~180 ms**, and LLM-first-sentence + TTS-TTFB is\n~377 ms. Output is **24 kHz mono**.\n\n## Prerequisites\n\n- **Hardware**: a single CUDA GPU. Reference run: 1x A10G 24 GB — ~6 GB for the\n  1.7B talker + ~8 GB for the codec decoder.\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\nMOSS-TTS-Realtime uses `model_type=moss_tts_realtime`, so its deploy config\n(`vllm_omni/deploy/moss_tts_realtime.yaml`, `codec_chunk_frames: 15` for low\nTTFA) is auto-discovered:\n\n```bash\nvllm serve OpenMOSS-Team/MOSS-TTS-Realtime --omni --port 8000\n```\n\n## Client usage\n\n### Streaming voice clone (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-Realtime\",\n    \"input\": \"This is a low-latency streaming TTS 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    \"stream\": true\n  }' --output output.wav\n```\n\n## Known limitations\n\n- Output is **24 kHz mono**.\n- Shares the `OpenMOSS-Team/MOSS-Audio-Tokenizer` codec (~7 GB, auto-downloaded;\n  override with `MOSS_TTS_CODEC_PATH`).\n\n## References\n\n- [Model card](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Realtime)\n- [MOSS-TTS Family](https://huggingface.co/OpenMOSS-Team/MOSS-TTS) · [GitHub](https://github.com/OpenMOSS/MOSS-TTS)\n",
  "source": {
    "source": "vllm-omni",
    "repo": "vllm-project/vllm-omni",
    "ref": "main",
    "sha": "cda4a1ef83b295ba332146a7709e6277725edc9f",
    "path": "recipes/models/OpenMOSS-Team/MOSS-TTS-Realtime.yaml",
    "edit_url": "https://github.com/vllm-project/vllm-omni/edit/main/recipes/models/OpenMOSS-Team/MOSS-TTS-Realtime.yaml",
    "overlays": []
  }
}