{
 "slug": "vibevoice",
 "name": "VibeVoice",
 "author": "Microsoft Research",
 "density": "full",
 "status": "active",
 "status_note": null,
 "alternatives_page": null,
 "one_liner": "Long-form multi-speaker TTS for podcast-style dialogue, up to 90 minutes and 4 speakers, in English and Chinese.",
 "category": {
  "slug": "voiceover",
  "name": "Voiceover (text-to-speech)"
 },
 "tags": [
  "Open source"
 ],
 "styles": [],
 "logo_url": "https://agentgid.com/media/avatars/microsoft.png",
 "logo_letter": "V",
 "links": {
  "repo": "https://github.com/microsoft/VibeVoice",
  "website": "https://huggingface.co/microsoft/VibeVoice-1.5B",
  "api_docs": null
 },
 "preview": null,
 "verdicts": [
  {
   "key": "price",
   "label": "Price",
   "value": "Free",
   "status": "yes",
   "sub": "No paid plan",
   "quote": null,
   "note": "Open weights, free to download.",
   "source": {
    "label": "huggingface.co/microsoft/VibeVoice-1.5B",
    "url": "https://huggingface.co/microsoft/VibeVoice-1.5B"
   },
   "n": 1
  },
  {
   "key": "commercial",
   "label": "Commercial use",
   "value": "With conditions",
   "status": "cond",
   "sub": "MIT",
   "quote": "The VibeVoice model is limited to research purpose use exploring highly realistic audio dialogue generation",
   "note": "The MIT license allows it, but Microsoft limits intended use to research and does not recommend commercial use.",
   "source": {
    "label": "huggingface.co/microsoft/VibeVoice-1.5B",
    "url": "https://huggingface.co/microsoft/VibeVoice-1.5B"
   },
   "n": 2
  },
  {
   "key": "credit",
   "label": "Credit the author",
   "value": "Required",
   "status": "cond",
   "sub": "Show a credit line",
   "quote": "The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.",
   "note": "Speech research made with Microsoft VibeVoice (MIT license)",
   "source": {
    "label": "LICENSE on GitHub",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/LICENSE"
   },
   "n": 3
  },
  {
   "key": "agent_access",
   "label": "Agent access",
   "value": "Download",
   "status": "info",
   "sub": "Files by hand",
   "quote": null,
   "note": "There is no package or API. Files are downloaded by hand from the official site.",
   "source": {
    "label": "github.com/microsoft/VibeVoice",
    "url": "https://github.com/microsoft/VibeVoice"
   },
   "n": 4
  }
 ],
 "license": {
  "name": "MIT (model card limits intended use to research)",
  "url": "https://huggingface.co/microsoft/VibeVoice-1.5B",
  "summary": "The code and model files are under the MIT license, so you keep the copyright notice with copies. But Microsoft says the model is for research only and does not recommend it in paid or real-world products.",
  "pages_read": 2,
  "rules": [
   "MIT-licensed weights, but Microsoft states the model is for research use and does not recommend commercial use.",
   "Every generated file carries an audible AI disclaimer and an imperceptible watermark.",
   "Not licensed for: voice cloning without recorded consent, disinformation, live deepfake voice conversion.",
   "Microsoft removed the TTS inference code from GitHub in September 2025; the 1.5B weights stay on Hugging Face."
  ],
  "rows": [
   {
    "label": "Commercial use",
    "value": "With conditions",
    "status": "cond",
    "quote": "The VibeVoice model is limited to research purpose use exploring highly realistic audio dialogue generation",
    "source": {
     "label": "huggingface.co/microsoft/VibeVoice-1.5B",
     "url": "https://huggingface.co/microsoft/VibeVoice-1.5B"
    },
    "note": null
   },
   {
    "label": "Credit",
    "value": "Required",
    "status": "cond",
    "quote": "The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.",
    "source": {
     "label": "LICENSE on GitHub",
     "url": "https://github.com/microsoft/VibeVoice/blob/main/LICENSE"
    },
    "note": null
   },
   {
    "label": "Inside your app or game",
    "value": "With conditions",
    "status": "cond",
    "quote": "We do not recommend using VibeVoice in commercial or real-world applications without further testing and development.",
    "source": {
     "label": "huggingface.co/microsoft/VibeVoice-1.5B",
     "url": "https://huggingface.co/microsoft/VibeVoice-1.5B"
    },
    "note": null
   },
   {
    "label": "Pass the files on to your users",
    "value": "With conditions",
    "status": "cond",
    "quote": "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software",
    "source": {
     "label": "LICENSE on GitHub",
     "url": "https://github.com/microsoft/VibeVoice/blob/main/LICENSE"
    },
    "note": "MIT lets you share copies if the copyright and permission notice go with them."
   },
   {
    "label": "AI training",
    "value": "Not stated",
    "status": "unknown",
    "quote": null,
    "source": null,
    "note": null
   },
   {
    "label": "Bots and bulk downloads",
    "value": "Not stated",
    "status": "unknown",
    "quote": null,
    "source": null,
    "note": null
   }
  ],
  "checked_at": "2026-10-25",
  "commercial_use_page": "https://agentgid.com/assets/vibevoice/commercial-use/"
 },
 "overview": {
  "heading": "VibeVoice: research voice model for long podcast-style dialogue",
  "body": "VibeVoice is a family of free voice models from Microsoft Research. The 1.5B model turns a written script into spoken conversation up to 90 minutes long with up to 4 voices, in English and Chinese. Microsoft limits it to research and removed the ready-to-run speech code, so it is for experiments, not for a product."
 },
 "uses": [
  {
   "title": "Research on podcast-style dialogue",
   "text": "Study how a model keeps several voices apart over a long script, for example a two-host talk show written as turns with speaker labels.",
   "idea": false,
   "image": null
  },
  {
   "title": "Long narration in one pass",
   "text": "Read a long text, such as a chapter or a lecture script, in one consistent voice instead of stitching many short clips together.",
   "idea": false,
   "image": null
  },
  {
   "title": "Let a chatbot start speaking early",
   "text": "With the small Realtime model, a chatbot answer can be read aloud while it is still being written, in English and with one voice.",
   "idea": false,
   "image": null
  },
  {
   "title": "Read a live data feed aloud",
   "text": "Turn a stream of short updates, such as scores or status messages, into speech for a research prototype.",
   "idea": false,
   "image": null
  },
  {
   "title": "Compare voice models for a decision",
   "text": "Listen to VibeVoice next to models with product-friendly licenses before you pick one for your app.",
   "idea": true,
   "image": null
  },
  {
   "title": "Test text clean-up before speech",
   "text": "Check how your app should clean text, such as numbers, symbols and code, before any voice model reads it.",
   "idea": true,
   "image": null
  }
 ],
 "start": {
  "prompt": "For a research test only, install microsoft/VibeVoice from GitHub with the streamingtts extra and read my text file with VibeVoice-Realtime-0.5B; do not use it in the product.",
  "install": null,
  "install_note": null,
  "code": [
   {
    "tab": "Step 1",
    "intro": "Use a machine with an NVIDIA graphics card (Microsoft names the T4) or a fast Apple Silicon Mac. Microsoft recommends its NVIDIA container for the setup.",
    "source": "sudo docker run --privileged --net=host --ipc=host --ulimit memlock=-1:-1 --ulimit stack=-1:-1 --gpus all --rm -it  nvcr.io/nvidia/pytorch:24.07-py3"
   },
   {
    "tab": "Step 2",
    "intro": "Get the code from GitHub and install it with the streaming extra. Only the Realtime 0.5B model has working install steps; the long-form 1.5B speech code was removed.",
    "source": "git clone https://github.com/microsoft/VibeVoice.git\ncd VibeVoice/\n\npip install -e .[streamingtts]"
   },
   {
    "tab": "Step 3",
    "intro": "Run the file example; the model downloads from Hugging Face on first run.",
    "source": "python demo/realtime_model_inference_from_file.py --model_path microsoft/VibeVoice-Realtime-0.5B --txt_path demo/text_examples/1p_vibevoice.txt --speaker_name Carter"
   }
  ]
 },
 "mistakes": [
  {
   "dont": "Looking for install steps or a script that runs the 1.5B long-form model from the official GitHub code",
   "fix": "Only the Realtime 0.5B model has official install and run steps; treat the 1.5B weights as research material",
   "why": "Microsoft removed the long-form speech code and marks its install section as disabled.",
   "dont_code": false,
   "fix_code": false
  },
  {
   "dont": "pip install -e . with the newest transformers for the Realtime model",
   "fix": "pip install -e .[streamingtts], which pins transformers 4.51.3",
   "why": "The streaming extra fixes the transformers version the Realtime model was built against.",
   "dont_code": true,
   "fix_code": true
  },
  {
   "dont": "Feeding raw chatbot output with code blocks, formulas or symbols straight into the Realtime model",
   "fix": "Strip or rewrite code, math and unusual symbols to plain words first",
   "why": "The model does not read code, formulas or uncommon symbols and gives unpredictable audio.",
   "dont_code": false,
   "fix_code": false
  },
  {
   "dont": "Sending one- to three-word replies such as \"OK, done.\" on their own",
   "fix": "Join very short replies with the next sentence before speaking them",
   "why": "Stability drops when the input is three words or fewer.",
   "dont_code": false,
   "fix_code": false
  },
  {
   "dont": "Using VibeVoice audio as the shipped voice of a product or a public video",
   "fix": "Use it for research, and pick a model with a product-friendly license for the shipped voice",
   "why": "Microsoft limits intended use to research, and every file carries an audible AI notice.",
   "dont_code": false,
   "fix_code": false
  }
 ],
 "agent": {
  "self_serve": false,
  "self_serve_text": null,
  "install": null,
  "repo": "https://github.com/microsoft/VibeVoice",
  "api_docs": null,
  "api_account": "Not needed. Everything runs on your side.",
  "get_files": null,
  "formats": [],
  "limits": null,
  "mcp": null,
  "mcp_url": null,
  "credit_line": "Speech research made with Microsoft VibeVoice (MIT license)",
  "rules_text": "- VibeVoice (Voiceover (text-to-speech)). Long-form multi-speaker TTS for podcast-style dialogue, up to 90 minutes and 4 speakers, in English and Chinese. Price: free; commercial use with conditions; credit required; ship inside your app with conditions.\n  Connect: download from https://huggingface.co/microsoft/VibeVoice-1.5B (no API).\n  Rule: MIT-licensed weights, but Microsoft states the model is for research use and does not recommend commercial use.\n  Rule: Every generated file carries an audible AI disclaimer and an imperceptible watermark.\n  Rule: Not licensed for: voice cloning without recorded consent, disinformation, live deepfake voice conversion.\n  Rule: Microsoft removed the TTS inference code from GitHub in September 2025; the 1.5B weights stay on Hugging Face.\n  License: https://huggingface.co/microsoft/VibeVoice-1.5B (checked 2026-10-25; details: https://agentgid.com/assets/vibevoice/)",
  "agent_md": "https://agentgid.com/assets/vibevoice/agent.md",
  "card_json": "https://agentgid.com/assets/vibevoice/card.json"
 },
 "popularity": {
  "gid_score": 52,
  "stars": 57295,
  "downloads_7d": null,
  "downloads_7d_label": null,
  "downloads_change_4w_pct": null,
  "eco": null,
  "package": null,
  "repo": "microsoft/VibeVoice",
  "agent_projects": null,
  "agent_projects_total": null,
  "downloads_unconfirmed": false,
  "series": {
   "downloads_weekly": [],
   "stars": [
    [
     "2026-11-07",
     56671
    ],
    [
     "2026-11-08",
     56728
    ],
    [
     "2026-11-09",
     56829
    ],
    [
     "2026-11-10",
     56935
    ],
    [
     "2026-11-11",
     56994
    ],
    [
     "2026-11-12",
     57062
    ],
    [
     "2026-11-13",
     57121
    ],
    [
     "2026-11-14",
     57163
    ],
    [
     "2026-11-15",
     57183
    ],
    [
     "2026-11-16",
     57199
    ],
    [
     "2026-11-17",
     57223
    ],
    [
     "2026-11-18",
     57247
    ],
    [
     "2026-11-19",
     57295
    ]
   ]
  },
  "stars_flat_since": null,
  "updated_at": "2026-11-19"
 },
 "alternatives": [
  {
   "slug": "vibevoice",
   "name": "VibeVoice",
   "why": null,
   "current": true,
   "price": "Free",
   "commercial": {
    "value": "With conditions",
    "status": "cond"
   },
   "access": "Download"
  },
  {
   "slug": "chatterbox",
   "name": "Chatterbox",
   "why": "Pick Chatterbox when you need an open voice model with an MIT license for a product, voice cloning and 23 languages.",
   "current": false,
   "price": "Free",
   "commercial": {
    "value": "Commercial OK",
    "status": "yes"
   },
   "access": "PyPI"
  },
  {
   "slug": "kokoro",
   "name": "Kokoro-82M",
   "why": "Pick Kokoro for a small open model under Apache-2.0 with fixed voices that you can ship in an app.",
   "current": false,
   "price": "Free",
   "commercial": {
    "value": "Commercial OK",
    "status": "yes"
   },
   "access": "npm, PyPI"
  },
  {
   "slug": "voicestudio",
   "name": "VoiceStudio",
   "why": "Pick VoiceStudio when you want a desktop app for voices, dubbing and transcription instead of running model code.",
   "current": false,
   "price": "Free",
   "commercial": {
    "value": "With conditions",
    "status": "cond"
   },
   "access": "MCP, API"
  },
  {
   "slug": "f5-tts",
   "name": "F5-TTS",
   "why": "Zero-shot voice-cloning TTS; code MIT, pre-trained weights CC BY-NC.",
   "current": false,
   "price": "Free",
   "commercial": {
    "value": "With conditions",
    "status": "cond"
   },
   "access": "PyPI"
  },
  {
   "slug": "mms-tts",
   "name": "Meta MMS-TTS",
   "why": "Family of small VITS text-to-speech checkpoints, one per language, covering over 1,100 languages.",
   "current": false,
   "price": "Free",
   "commercial": {
    "value": "Non-commercial",
    "status": "no"
   },
   "access": "Download"
  }
 ],
 "faq": [
  {
   "q": "Can I use VibeVoice for the voice in my paid app?",
   "a": "Microsoft does not recommend it. The license is MIT, but the model card limits the model to research and says it is not for commercial or real-world use without more work."
  },
  {
   "q": "Can I still run the 90-minute model?",
   "a": "The weights are still on Hugging Face, but Microsoft removed the speech code from GitHub and marks its install steps as disabled."
  },
  {
   "q": "How many voices can talk in one file?",
   "a": "Up to 4 with the 1.5B model. The Realtime 0.5B model has one voice at a time."
  },
  {
   "q": "Which languages work?",
   "a": "The 1.5B model is trained on English and Chinese only; the Realtime model is meant for English, with nine more languages marked as experimental."
  }
 ],
 "at_a_glance": {
  "latest_version": "1.0.0",
  "last_commit": "2026-10-08",
  "runtime": "Python 3.10+; the streamingtts extra pins transformers 4.51.3; an NVIDIA GPU (T4 or better) or an Apple M4 Pro class Mac for real-time speed.",
  "key_needed": false
 },
 "good_to_know": [
  "Every file it makes carries a spoken AI notice and a hidden watermark. You cannot get clean voice audio for a product this way.",
  "Copying a real person's voice without recorded consent, fake recordings and live voice changing are outside what the release allows.",
  "The long-form 1.5B speech model has no official way to run it now. Its install steps say they were turned off after misuse.",
  "It speaks only English and Chinese well, and only speech. It does not make music or sound effects on purpose, and it can add background music by chance.",
  "The small Realtime model has one voice at a time and English only. Voices come built in; for your own voice you must contact Microsoft."
 ],
 "stack_snippets": [],
 "claims": [
  "Up to 90 minutes of speech with up to 4 distinct speakers",
  "The small Realtime model gives first audible speech in about 300 ms (hardware dependent)"
 ],
 "checked": {
  "license_read": "2026-10-25",
  "overview_checked": "2026-11-10",
  "numbers_updated": "2026-11-19",
  "sources": [
   {
    "label": "huggingface.co/microsoft/VibeVoice-1.5B",
    "url": "https://huggingface.co/microsoft/VibeVoice-1.5B"
   },
   {
    "label": "huggingface.co",
    "url": "https://huggingface.co/microsoft/VibeVoice-Realtime-0.5B"
   },
   {
    "label": "github.com/microsoft/VibeVoice",
    "url": "https://github.com/microsoft/VibeVoice"
   },
   {
    "label": "README on GitHub",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/README.md"
   },
   {
    "label": "github.com/microsoft/VibeVoice/blob/main/docs/vibevoice-tts.md",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/docs/vibevoice-tts.md"
   },
   {
    "label": "github.com/microsoft/VibeVoice/blob/main/docs/vibevoice-realtime-0.5b.md",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/docs/vibevoice-realtime-0.5b.md"
   },
   {
    "label": "LICENSE on GitHub",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/LICENSE"
   },
   {
    "label": "github.com/microsoft/VibeVoice/blob/main/pyproject.toml",
    "url": "https://github.com/microsoft/VibeVoice/blob/main/pyproject.toml"
   },
   {
    "label": "huggingface.co",
    "url": "https://huggingface.co/api/models/microsoft/VibeVoice-1.5B"
   },
   {
    "label": "pypi.org/pypi/vibevoice/json",
    "url": "https://pypi.org/pypi/vibevoice/json"
   }
  ],
  "unverified": "The license text (MIT) allows commercial use, while the model card restricts intended use to research; commercial and in_app set to 'conditions' for that conflict."
 },
 "user_reports": null,
 "url": "https://agentgid.com/assets/vibevoice/",
 "note": "Written by AgentGid from the official pages in checked.sources; quotes are copied from those pages. status: yes | cond | no | unknown | info. Prices and vendor claims are as reported on the checked date. Not legal advice."
}
