{
 "generated": "2026-10-15",
 "about": "Open-weight text-to-speech and speech-to-text models you can run on your own computer; memory is approximate (stated on the card, or derived from file size as noted), and release_date is the card's stated date, else the Hugging Face repo creation date.",
 "leaderboards": [
  {
   "name": "Open ASR Leaderboard (English)",
   "task": "stt",
   "metric": "Average WER %",
   "lower_is_better": true,
   "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
   "retrieved": "2026-10-15",
   "best_closed": {
    "name": "OtterAI speech-v3.1",
    "value": 4.55
   }
  },
  {
   "name": "Artificial Analysis Speech Arena",
   "task": "tts",
   "metric": "Elo",
   "lower_is_better": false,
   "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
   "retrieved": "2026-10-15",
   "best_closed": {
    "name": "Eleven v4 (ElevenLabs)",
    "value": 1330
   }
  },
  {
   "name": "TTS Arena V2",
   "task": "tts",
   "metric": "Elo",
   "lower_is_better": false,
   "url": "https://huggingface.co/spaces/TTS-AGI/TTS-Arena-V2",
   "retrieved": "2026-10-15",
   "best_closed": {
    "name": "Inworld TTS MAX",
    "value": 1553
   }
  }
 ],
 "models": [
  {
   "id": "qwen3-asr-1-7b",
   "name": "Qwen3-ASR 1.7B",
   "maker": "Alibaba Qwen",
   "task": "stt",
   "params_m": 2349,
   "memory_gb": 3,
   "languages": "30 languages + 22 Chinese dialects",
   "voice_cloning": null,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 5.4,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "qwen-asr"
   ],
   "mac_ok": true,
   "release_date": "2026-01-28",
   "hf": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
   "best_for": "Most accurate open transcription right now",
   "notes": "Memory ~1.2x the 2.5 GB MLX 8-bit file; streaming via the vLLM backend.",
   "sources": [
    {
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "cohere-transcribe",
   "name": "Cohere Transcribe (03-2026)",
   "maker": "Cohere",
   "task": "stt",
   "params_m": 2066,
   "memory_gb": 5,
   "languages": "14 languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 5.77,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "transcribe.cpp"
   ],
   "mac_ok": true,
   "release_date": "2026-03-26",
   "hf": "https://huggingface.co/CohereLabs/cohere-transcribe-03-2026",
   "best_for": "Accurate dictation in 14 major languages",
   "notes": "Memory ~1.2x the 4.1 GB BF16 weights; no timestamps, diarization or auto language detection.",
   "sources": [
    {
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-03-2026",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "canary-qwen-2-5b",
   "name": "Canary-Qwen 2.5B",
   "maker": "NVIDIA",
   "task": "stt",
   "params_m": 2559,
   "memory_gb": 6,
   "languages": "English only",
   "voice_cloning": null,
   "realtime": null,
   "license": "CC-BY-4.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 5.94,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "nemo-parakeet"
   ],
   "mac_ok": null,
   "release_date": "2025-07-17",
   "hf": "https://huggingface.co/nvidia/canary-qwen-2.5b",
   "best_for": "High-accuracy English on an NVIDIA GPU",
   "notes": "Memory ~1.2x the 5.1 GB weights; runs through NVIDIA NeMo.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/canary-qwen-2.5b",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "granite-speech-4-1-2b",
   "name": "Granite Speech 4.1 2B",
   "maker": "IBM",
   "task": "stt",
   "params_m": 2313,
   "memory_gb": 6,
   "languages": "EN, FR, DE, ES, PT, JA",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 5.96,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-04-29",
   "hf": "https://huggingface.co/ibm-granite/granite-speech-4.1-2b",
   "best_for": "Transcribe and translate, business-friendly license",
   "notes": "Memory ~1.2x the 4.8 GB weights; also translates speech to and from English.",
   "sources": [
    {
     "url": "https://huggingface.co/ibm-granite/granite-speech-4.1-2b",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "parakeet-tdt-0-6b-v2",
   "name": "Parakeet TDT 0.6B v2",
   "maker": "NVIDIA",
   "task": "stt",
   "params_m": 600,
   "memory_gb": 3,
   "languages": "English only",
   "voice_cloning": null,
   "realtime": null,
   "license": "CC-BY-4.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.02,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "parakeet-mlx",
    "mlx-audio",
    "nemo-parakeet"
   ],
   "mac_ok": true,
   "release_date": "2025-05-01",
   "hf": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2",
   "best_for": "Very fast English with word timestamps",
   "notes": "Memory ~1.2x the 2.5 GB .nemo file.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v2",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "moss-transcribe-diarize",
   "name": "MOSS-Transcribe-Diarize 0.9B",
   "maker": "OpenMOSS",
   "task": "stt",
   "params_m": 909,
   "memory_gb": 2.5,
   "languages": "50+ languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.17,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-07-09",
   "hf": "https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize",
   "best_for": "Meetings: who said what, and when",
   "notes": "Memory ~1.3x the 1.8 GB weights; handles up to 90 minutes in one pass.",
   "sources": [
    {
     "url": "https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "parakeet-tdt-0-6b-v3",
   "name": "Parakeet TDT 0.6B v3",
   "maker": "NVIDIA",
   "task": "stt",
   "params_m": 600,
   "memory_gb": 3,
   "languages": "25 European languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "CC-BY-4.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.18,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "parakeet-mlx",
    "mlx-audio",
    "handy",
    "nemo-parakeet"
   ],
   "mac_ok": true,
   "release_date": "2025-08-14",
   "hf": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3",
   "best_for": "Fast offline dictation on a Mac",
   "notes": "Memory ~1.2x the 2.5 GB weights; a 0.7 GB Q8 GGUF is also published.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/senstella/parakeet-mlx",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/cjpais/Handy",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "qwen3-asr-0-6b",
   "name": "Qwen3-ASR 0.6B",
   "maker": "Alibaba Qwen",
   "task": "stt",
   "params_m": 938,
   "memory_gb": 1,
   "languages": "30 languages + 22 Chinese dialects",
   "voice_cloning": null,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.24,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "qwen-asr"
   ],
   "mac_ok": true,
   "release_date": "2026-01-28",
   "hf": "https://huggingface.co/Qwen/Qwen3-ASR-0.6B",
   "best_for": "Small multilingual model for lighter machines",
   "notes": "Memory ~1.4x the 0.7 GB MLX 4-bit file.",
   "sources": [
    {
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-0.6B",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/Qwen3-ASR-0.6B-4bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "granite-speech-5-0-turboctc",
   "name": "Granite Speech 5.0 470M TurboCTC",
   "maker": "IBM",
   "task": "stt",
   "params_m": 473,
   "memory_gb": 1,
   "languages": "English only",
   "voice_cloning": null,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.38,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-08-25",
   "hf": "https://huggingface.co/ibm-granite/granite-speech-5.0-470m-turboctc",
   "best_for": "Extremely fast English on laptops and phones",
   "notes": "Memory ~1.1x the 0.95 GB weights; built for edge devices.",
   "sources": [
    {
     "url": "https://huggingface.co/ibm-granite/granite-speech-5.0-470m-turboctc",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "distil-large-v3-5",
   "name": "Distil-Whisper large-v3.5",
   "maker": "Hugging Face (distil-whisper)",
   "task": "stt",
   "params_m": 756,
   "memory_gb": 2,
   "languages": "English only",
   "voice_cloning": null,
   "realtime": null,
   "license": "MIT",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 6.76,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "whisper-cpp",
    "faster-whisper",
    "whisper",
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2024-12-05",
   "hf": "https://huggingface.co/distil-whisper/distil-large-v3.5",
   "best_for": "Faster drop-in Whisper for English",
   "notes": "Memory ~1.3x the FP16 size (half the 3.0 GB FP32 file).",
   "sources": [
    {
     "url": "https://huggingface.co/distil-whisper/distil-large-v3.5",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "whisper-large-v3",
   "name": "Whisper large-v3",
   "maker": "OpenAI",
   "task": "stt",
   "params_m": 1543,
   "memory_gb": 3.9,
   "languages": "99+ languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 7.16,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "whisper-cpp",
    "whisperkit",
    "faster-whisper",
    "whisper",
    "handy"
   ],
   "mac_ok": true,
   "release_date": "2023-11-07",
   "hf": "https://huggingface.co/openai/whisper-large-v3",
   "best_for": "The safe multilingual default",
   "notes": "Memory from whisper.cpp's table (~3.9 GB for large); PyTorch needs about 10 GB.",
   "sources": [
    {
     "url": "https://huggingface.co/openai/whisper-large-v3",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/ggml-org/whisper.cpp",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/openai/whisper",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/argmaxinc/whisperkit-coreml",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "voxtral-mini-3b",
   "name": "Voxtral Mini 3B (2507)",
   "maker": "Mistral AI",
   "task": "stt",
   "params_m": 4676,
   "memory_gb": 9.5,
   "languages": "8 languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 7.25,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "vLLM"
   ],
   "mac_ok": true,
   "release_date": null,
   "hf": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507",
   "best_for": "Transcribe, then ask questions about audio",
   "notes": "Memory from the card (~9.5 GB GPU in BF16); also a capable text chat model.",
   "sources": [
    {
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "canary-1b-v2",
   "name": "Canary 1B v2",
   "maker": "NVIDIA",
   "task": "stt",
   "params_m": 979,
   "memory_gb": 4.5,
   "languages": "25 European languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "CC-BY-4.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 7.27,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "nemo-parakeet"
   ],
   "mac_ok": true,
   "release_date": "2025-08-14",
   "hf": "https://huggingface.co/nvidia/canary-1b-v2",
   "best_for": "European languages plus translation to English",
   "notes": "Memory ~1.15x the 3.9 GB FP32 weights.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/canary-1b-v2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio/tree/main/mlx_audio/stt/models/canary",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "whisper-large-v3-turbo",
   "name": "Whisper large-v3-turbo",
   "maker": "OpenAI",
   "task": "stt",
   "params_m": 809,
   "memory_gb": 2,
   "languages": "99+ languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "MIT",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 7.64,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "whisper-cpp",
    "whisperkit",
    "mlx-audio",
    "faster-whisper",
    "whisper",
    "handy"
   ],
   "mac_ok": true,
   "release_date": "2024-10-01",
   "hf": "https://huggingface.co/openai/whisper-large-v3-turbo",
   "best_for": "Fast multilingual dictation, runs everywhere",
   "notes": "Memory ~1.25x the 1.6 GB ggml file Handy downloads.",
   "sources": [
    {
     "url": "https://huggingface.co/openai/whisper-large-v3-turbo",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/argmaxinc/argmax-oss-swift",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/cjpais/Handy",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "vibevoice-asr",
   "name": "VibeVoice-ASR",
   "maker": "Microsoft",
   "task": "stt",
   "params_m": 8674,
   "memory_gb": 11,
   "languages": "50+ languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "MIT",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 8.16,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-01-21",
   "hf": "https://huggingface.co/microsoft/VibeVoice-ASR",
   "best_for": "Hour-long recordings with speakers and timestamps",
   "notes": "Memory ~1.15x the 9.5 GB MLX 8-bit file.",
   "sources": [
    {
     "url": "https://huggingface.co/microsoft/VibeVoice-ASR",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "voxtral-mini-4b-realtime",
   "name": "Voxtral Mini 4B Realtime (2602)",
   "maker": "Mistral AI",
   "task": "stt",
   "params_m": 4430,
   "memory_gb": 4,
   "languages": "13 languages",
   "voice_cloning": null,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 8.44,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "vLLM"
   ],
   "mac_ok": true,
   "release_date": "2026-02-04",
   "hf": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602",
   "best_for": "Live captions and voice assistants",
   "notes": "Memory ~1.3x the 3.1 GB MLX 4-bit file; delay configurable from 240 ms.",
   "sources": [
    {
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://mistral.ai/news/voxtral-transcribe-2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "nemotron-3-5-asr-streaming",
   "name": "Nemotron 3.5 ASR Streaming 0.6B",
   "maker": "NVIDIA",
   "task": "stt",
   "params_m": 638,
   "memory_gb": 1.5,
   "languages": "40 language-locales",
   "voice_cloning": null,
   "realtime": true,
   "license": "OpenMDW-1.1",
   "commercial_ok": true,
   "quality": {
    "board": "Open ASR Leaderboard (English)",
    "value": 9.75,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "nemo-parakeet"
   ],
   "mac_ok": true,
   "release_date": "2026-06-04",
   "hf": "https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b",
   "best_for": "Low-latency multilingual streaming for voice agents",
   "notes": "Memory ~1.2x the 1.3 GB MLX file.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/spaces/hf-audio/open_asr_leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "moonshine-streaming-medium",
   "name": "Moonshine Streaming Medium",
   "maker": "Moonshine AI (Useful Sensors)",
   "task": "stt",
   "params_m": 245,
   "memory_gb": 1,
   "languages": "English only",
   "voice_cloning": null,
   "realtime": true,
   "license": "MIT",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "mlx-audio",
    "Moonshine Voice"
   ],
   "mac_ok": true,
   "release_date": "2026-01-06",
   "hf": "https://huggingface.co/moonshine-ai/moonshine-streaming-medium",
   "best_for": "Live English captions on small devices",
   "notes": "Memory ~1x the 1.06 GB FP32 weights; tiny (34M) and small (123M) sizes also exist.",
   "sources": [
    {
     "url": "https://huggingface.co/moonshine-ai/moonshine-streaming-medium",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "whisper-small",
   "name": "Whisper small",
   "maker": "OpenAI",
   "task": "stt",
   "params_m": 244,
   "memory_gb": 0.9,
   "languages": "99+ languages",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "whisper-cpp",
    "whisperkit",
    "faster-whisper",
    "whisper",
    "handy"
   ],
   "mac_ok": true,
   "release_date": "2022-09-26",
   "hf": "https://huggingface.co/openai/whisper-small",
   "best_for": "Light dictation on older or low-RAM machines",
   "notes": "Memory from whisper.cpp's table (~0.85 GB for small).",
   "sources": [
    {
     "url": "https://huggingface.co/openai/whisper-small",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/ggml-org/whisper.cpp",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/cjpais/Handy",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/argmaxinc/whisperkit-coreml",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "fish-audio-drama-3",
   "name": "Fish Audio Drama 3 (preview)",
   "maker": "Fish Audio",
   "task": "tts",
   "params_m": 35900,
   "memory_gb": 28,
   "languages": "83 languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "Fish Audio Research License",
   "commercial_ok": null,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1241,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "fish-speech",
    "SGLang-Omni"
   ],
   "mac_ok": true,
   "release_date": "2026-10-01",
   "hf": "https://huggingface.co/fishaudio/drama-3",
   "best_for": "Best open voice quality, if you have RAM",
   "notes": "35B MoE (3B active); memory ~1.2x the 23.8 GB MLX 4-bit file; free commercial use only under a revenue cap.",
   "sources": [
    {
     "url": "https://huggingface.co/fishaudio/drama-3",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/fish-audio-drama-3-4bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "breeze-tts-2",
   "name": "Breeze TTS 2",
   "maker": "BreezeBlue",
   "task": "tts",
   "params_m": 3466,
   "memory_gb": 8,
   "languages": "EN, ZH",
   "voice_cloning": true,
   "realtime": true,
   "license": "BreezeBlue Research & Non-Commercial",
   "commercial_ok": false,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1221,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-08-25",
   "hf": "https://huggingface.co/BreezeBlue/Breeze-TTS-2",
   "best_for": "Top open voice cloning for personal projects",
   "notes": "Memory from the card (about 7.7 GiB GPU, eager mode); official code targets NVIDIA, Mac via MLX-Audio.",
   "sources": [
    {
     "url": "https://huggingface.co/BreezeBlue/Breeze-TTS-2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio/tree/main/mlx_audio/tts/models",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "fish-audio-s2-pro",
   "name": "Fish Audio S2 Pro",
   "maker": "Fish Audio",
   "task": "tts",
   "params_m": 4562,
   "memory_gb": 8,
   "languages": "80+ languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "Fish Audio Research License",
   "commercial_ok": false,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1118,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "fish-speech",
    "mlx-audio",
    "vLLM-Omni"
   ],
   "mac_ok": true,
   "release_date": "2026-03-09",
   "hf": "https://huggingface.co/fishaudio/s2-pro",
   "best_for": "Expressive multilingual speech with inline emotion tags",
   "notes": "Memory ~1.2x the 6.7 GB MLX 8-bit file; commercial use needs a paid license.",
   "sources": [
    {
     "url": "https://huggingface.co/fishaudio/s2-pro",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/fishaudio/fish-speech",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/fish-audio-s2-pro-8bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "step-audio-editx",
   "name": "Step-Audio-EditX",
   "maker": "StepFun",
   "task": "tts",
   "params_m": 3530,
   "memory_gb": 12,
   "languages": "ZH, EN, JA, KO + Sichuanese, Cantonese",
   "voice_cloning": true,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1096,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "Step-Audio-EditX code"
   ],
   "mac_ok": null,
   "release_date": "2025-11-12",
   "hf": "https://huggingface.co/stepfun-ai/Step-Audio-EditX",
   "best_for": "Editing emotion and style of speech",
   "notes": "Memory from the card (12 GB GPU, 16 GB safer); official code needs an NVIDIA GPU.",
   "sources": [
    {
     "url": "https://huggingface.co/stepfun-ai/Step-Audio-EditX",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/stepfun-ai/Step-Audio-EditX",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "voxtral-tts",
   "name": "Voxtral TTS (4B)",
   "maker": "Mistral AI",
   "task": "tts",
   "params_m": 4000,
   "memory_gb": 16,
   "languages": "9 languages",
   "voice_cloning": null,
   "realtime": true,
   "license": "CC-BY-NC-4.0",
   "commercial_ok": false,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1088,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "vLLM-Omni"
   ],
   "mac_ok": true,
   "release_date": "2026-03-26",
   "hf": "https://huggingface.co/mistralai/Voxtral-4B-TTS-2603",
   "best_for": "Low-latency voice agents in European languages",
   "notes": "Memory from the card (16 GB GPU, BF16); open release ships 20 preset voices, custom voices via Mistral.",
   "sources": [
    {
     "url": "https://huggingface.co/mistralai/Voxtral-4B-TTS-2603",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://siliconangle.com/2026/03/26/mistral-releases-open-weights-speaking-AI-model-voxtral-tts/",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "magpie-tts-multilingual",
   "name": "Magpie TTS Multilingual 357M",
   "maker": "NVIDIA",
   "task": "tts",
   "params_m": 364,
   "memory_gb": 2,
   "languages": "12 languages",
   "voice_cloning": false,
   "realtime": null,
   "license": "NVIDIA Open Model License",
   "commercial_ok": true,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1063,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "nemo-parakeet"
   ],
   "mac_ok": null,
   "release_date": null,
   "hf": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m",
   "best_for": "Commercial multilingual voices for voice agents",
   "notes": "Memory ~1.3x the 1.47 GB .nemo file; fixed voices, no cloning.",
   "sources": [
    {
     "url": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "kokoro-82m",
   "name": "Kokoro 82M v1.0",
   "maker": "hexgrad",
   "task": "tts",
   "params_m": 82,
   "memory_gb": 0.5,
   "languages": "EN, JA, ZH, FR, ES, IT, PT, HI",
   "voice_cloning": false,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1061,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2025-01-27",
   "hf": "https://huggingface.co/hexgrad/Kokoro-82M",
   "best_for": "Fast, tiny, free for commercial use",
   "notes": "Memory ~1.5x the 0.33 GB weights file; 54 preset voices.",
   "sources": [
    {
     "url": "https://huggingface.co/hexgrad/Kokoro-82M",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "maya1",
   "name": "Maya1",
   "maker": "Maya Research",
   "task": "tts",
   "params_m": 3301,
   "memory_gb": 8,
   "languages": null,
   "voice_cloning": false,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1045,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2025-10-18",
   "hf": "https://huggingface.co/maya-research/maya1",
   "best_for": "Design a voice by describing it",
   "notes": "Memory ~1.2x the 6.6 GB BF16 weights; voices come from text descriptions, not reference audio.",
   "sources": [
    {
     "url": "https://huggingface.co/maya-research/maya1",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/maya1-4bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "higgs-tts-3",
   "name": "Higgs TTS 3 (4B)",
   "maker": "Boson AI",
   "task": "tts",
   "params_m": 4655,
   "memory_gb": 11,
   "languages": "100+ languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "Boson Higgs TTS 3 Research & Non-Commercial",
   "commercial_ok": false,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1037,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "mlx-audio",
    "SGLang-Omni"
   ],
   "mac_ok": true,
   "release_date": "2026-06-04",
   "hf": "https://huggingface.co/bosonai/higgs-tts-3-4b",
   "best_for": "Many languages; creators may monetize content",
   "notes": "Memory ~1.2x the 9.3 GB BF16 weights; products need a paid license, creator content is free.",
   "sources": [
    {
     "url": "https://huggingface.co/bosonai/higgs-tts-3-4b",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "chatterbox",
   "name": "Chatterbox (Multilingual V3)",
   "maker": "Resemble AI",
   "task": "tts",
   "params_m": 500,
   "memory_gb": 3.5,
   "languages": "23 languages",
   "voice_cloning": true,
   "realtime": null,
   "license": "MIT",
   "commercial_ok": true,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 1025,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "chatterbox",
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2025-04-24",
   "hf": "https://huggingface.co/ResembleAI/chatterbox",
   "best_for": "Free voice cloning with commercial rights",
   "notes": "Memory ~1.3x the 2.7 GB MLX V3 files; the arena score may be for the original English model.",
   "sources": [
    {
     "url": "https://huggingface.co/ResembleAI/chatterbox",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "vibevoice-1-5b",
   "name": "VibeVoice 1.5B",
   "maker": "Microsoft",
   "task": "tts",
   "params_m": 2704,
   "memory_gb": 6.5,
   "languages": "EN, ZH",
   "voice_cloning": null,
   "realtime": null,
   "license": "MIT",
   "commercial_ok": null,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 956,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "VibeVoice code"
   ],
   "mac_ok": null,
   "release_date": "2025-08-25",
   "hf": "https://huggingface.co/microsoft/VibeVoice-1.5B",
   "best_for": "Long podcasts with up to four speakers",
   "notes": "Memory ~1.2x the 5.4 GB BF16 weights; Microsoft says research use only.",
   "sources": [
    {
     "url": "https://huggingface.co/microsoft/VibeVoice-1.5B",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "xtts-v2",
   "name": "XTTS v2",
   "maker": "Coqui",
   "task": "tts",
   "params_m": null,
   "memory_gb": 2.5,
   "languages": "17 languages",
   "voice_cloning": true,
   "realtime": null,
   "license": "Coqui Public Model License",
   "commercial_ok": false,
   "quality": {
    "board": "Artificial Analysis Speech Arena",
    "value": 914,
    "as_of": "2026-10-15"
   },
   "runs_with": [
    "Coqui TTS"
   ],
   "mac_ok": null,
   "release_date": "2023-10-31",
   "hf": "https://huggingface.co/coqui/XTTS-v2",
   "best_for": "Classic cloning from a 6-second clip",
   "notes": "Memory ~1.2x the 2.1 GB checkpoint files; older model, license is non-commercial.",
   "sources": [
    {
     "url": "https://huggingface.co/coqui/XTTS-v2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://artificialanalysis.ai/text-to-speech/leaderboard",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "qwen3-tts-1-7b",
   "name": "Qwen3-TTS 1.7B",
   "maker": "Alibaba Qwen",
   "task": "tts",
   "params_m": 1929,
   "memory_gb": 4,
   "languages": "10 languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "mlx-audio",
    "whisperkit",
    "qwen-tts"
   ],
   "mac_ok": true,
   "release_date": "2026-01-21",
   "hf": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base",
   "best_for": "Cloning, preset voices and voice design",
   "notes": "Memory ~1.3x the 3.1 GB MLX 8-bit file; WhisperKit's package runs it via TTSKit.",
   "sources": [
    {
     "url": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/argmaxinc/argmax-oss-swift",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "omnivoice",
   "name": "OmniVoice",
   "maker": "k2-fsa",
   "task": "tts",
   "params_m": 613,
   "memory_gb": 2,
   "languages": "600+ languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "CC-BY-NC (weights)",
   "commercial_ok": false,
   "quality": null,
   "runs_with": [
    "mlx-audio",
    "omnivoice"
   ],
   "mac_ok": true,
   "release_date": "2026-03-30",
   "hf": "https://huggingface.co/k2-fsa/OmniVoice",
   "best_for": "Rare languages and fast cloning",
   "notes": "Memory ~1.4x the 1.4 GB MLX 8-bit file; code Apache-2.0 but weights non-commercial.",
   "sources": [
    {
     "url": "https://huggingface.co/k2-fsa/OmniVoice",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "voxcpm2",
   "name": "VoxCPM2",
   "maker": "OpenBMB",
   "task": "tts",
   "params_m": 2290,
   "memory_gb": 4,
   "languages": "30 languages",
   "voice_cloning": true,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-04-03",
   "hf": "https://huggingface.co/openbmb/VoxCPM2",
   "best_for": "48 kHz cloning, free for commercial use",
   "notes": "Memory ~1.2x the 3.2 GB MLX 8-bit file.",
   "sources": [
    {
     "url": "https://huggingface.co/openbmb/VoxCPM2",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "fun-cosyvoice3-0-5b",
   "name": "Fun-CosyVoice 3 0.5B",
   "maker": "Alibaba FunAudioLLM",
   "task": "tts",
   "params_m": 500,
   "memory_gb": 2,
   "languages": "9 languages + 18 Chinese dialects",
   "voice_cloning": true,
   "realtime": true,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "cosyvoice",
    "mlx-audio-plus"
   ],
   "mac_ok": true,
   "release_date": "2025-12-11",
   "hf": "https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512",
   "best_for": "Chinese dialects and low-latency streaming",
   "notes": "Memory ~1.4x the 1.4 GB MLX 8-bit file (mlx-audio-plus).",
   "sources": [
    {
     "url": "https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/Fun-CosyVoice3-0.5B-2512-8bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "f5-tts-v1",
   "name": "F5-TTS v1",
   "maker": "SWivid",
   "task": "tts",
   "params_m": null,
   "memory_gb": 2,
   "languages": null,
   "voice_cloning": true,
   "realtime": null,
   "license": "CC-BY-NC-4.0",
   "commercial_ok": false,
   "quality": null,
   "runs_with": [
    "F5-TTS",
    "f5-tts-mlx"
   ],
   "mac_ok": true,
   "release_date": "2025-03-12",
   "hf": "https://huggingface.co/SWivid/F5-TTS",
   "best_for": "Simple zero-shot cloning for hobby projects",
   "notes": "Memory ~1.5x the 1.35 GB checkpoint; code MIT, weights non-commercial.",
   "sources": [
    {
     "url": "https://huggingface.co/SWivid/F5-TTS",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/SWivid/F5-TTS",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "indextts-2-5",
   "name": "IndexTTS-2.5",
   "maker": "bilibili Index Team",
   "task": "tts",
   "params_m": null,
   "memory_gb": 6,
   "languages": "ZH, EN, JA, ES, AR",
   "voice_cloning": true,
   "realtime": null,
   "license": "bilibili Model License",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "index-tts"
   ],
   "mac_ok": false,
   "release_date": "2026-08-10",
   "hf": "https://huggingface.co/IndexTeam/IndexTTS-2.5",
   "best_for": "Cloning with emotion control",
   "notes": "Memory from the card (~6 GB VRAM, NVIDIA GPU required); companies over 100M users need a license.",
   "sources": [
    {
     "url": "https://huggingface.co/IndexTeam/IndexTTS-2.5",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/IndexTeam/IndexTTS-2.5/blob/main/LICENSE",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "dia-1-6b",
   "name": "Dia 1.6B",
   "maker": "Nari Labs",
   "task": "tts",
   "params_m": 1611,
   "memory_gb": 10,
   "languages": "English only",
   "voice_cloning": true,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "dia",
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2025-04-20",
   "hf": "https://huggingface.co/nari-labs/Dia-1.6B",
   "best_for": "Two-speaker dialogue with laughs and sighs",
   "notes": "Memory from the card (~10 GB VRAM full precision); MLX 4-bit file is 3.2 GB.",
   "sources": [
    {
     "url": "https://huggingface.co/nari-labs/Dia-1.6B",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/Dia-1.6B-4bit",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "kitten-tts-0-8",
   "name": "Kitten TTS 0.8 nano",
   "maker": "KittenML",
   "task": "tts",
   "params_m": 15,
   "memory_gb": 0.1,
   "languages": "English",
   "voice_cloning": null,
   "realtime": null,
   "license": "Apache-2.0",
   "commercial_ok": true,
   "quality": null,
   "runs_with": [
    "KittenTTS",
    "mlx-audio"
   ],
   "mac_ok": true,
   "release_date": "2026-02-19",
   "hf": "https://huggingface.co/KittenML/kitten-tts-nano-0.8-fp32",
   "best_for": "Tiny CPU-only voice for any machine",
   "notes": "Memory ~2x the 56 MB file; micro (40M) and mini (80M) sizes also exist.",
   "sources": [
    {
     "url": "https://huggingface.co/KittenML/kitten-tts-nano-0.8-fp32",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/KittenML/KittenTTS",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://github.com/Blaizzy/mlx-audio",
     "retrieved": "2026-10-15"
    }
   ]
  },
  {
   "id": "vibevoice-realtime-0-5b",
   "name": "VibeVoice Realtime 0.5B",
   "maker": "Microsoft",
   "task": "tts",
   "params_m": 1018,
   "memory_gb": 1.5,
   "languages": "English (+9 experimental)",
   "voice_cloning": null,
   "realtime": true,
   "license": "MIT",
   "commercial_ok": null,
   "quality": null,
   "runs_with": [
    "mlx-audio",
    "VibeVoice code"
   ],
   "mac_ok": true,
   "release_date": "2025-12-04",
   "hf": "https://huggingface.co/microsoft/VibeVoice-Realtime-0.5B",
   "best_for": "Streaming speech as text arrives",
   "notes": "Memory ~1.2x the 1.2 GB MLX 8-bit file; Microsoft limits it to research use.",
   "sources": [
    {
     "url": "https://huggingface.co/microsoft/VibeVoice-Realtime-0.5B",
     "retrieved": "2026-10-15"
    },
    {
     "url": "https://huggingface.co/mlx-community/VibeVoice-Realtime-0.5B-8bit",
     "retrieved": "2026-10-15"
    }
   ]
  }
 ]
}