The most recognized open text-to-speech and voice-cloning models

The open TTS and voice-cloning models you can run yourself — Kokoro, Chatterbox, XTTS-v2, F5-TTS and more. Ranked by recognition, each with the honest VRAM-to-run, license, and runner. We list the recognized open voice models with transparent costs (many run on a small GPU or even CPU); which voice you prefer is yours to judge.

How this is ranked: Objective task filter over the audio catalog, ordered by notability/recognition. We don't rank naturalness/quality — we present the recognized open voice models with honest run-costs and flag non-commercial licenses (XTTS-v2, F5-TTS are non-commercial).

1. Qwen2.5-Omni-7B — Omni understanding + speech generation, Alibaba Qwen · ~24 GB VRAM · commercial OK
2. Kokoro-82M — Text → speech, hexgrad · ~1.0 GB VRAM · commercial OK
3. XTTS-v2 — Voice cloning, Coqui · ~6.0 GB VRAM · non-commercial
4. Chatterbox — Text → speech, Resemble AI · ~8.0 GB VRAM · commercial OK
5. MiniCPM-o 2.6 — Omni understanding + speech generation, OpenBMB · ~9.0 GB VRAM · commercial OK
6. F5-TTS — Voice cloning, SWivid · ~6.0 GB VRAM · non-commercial
7. Dia 1.6B — Text → speech, Nari Labs · ~10 GB VRAM · commercial OK
8. Bark — Text → speech, Suno · ~8.0 GB VRAM · commercial OK
9. Ming-Lite-Omni — Omni understanding + image & speech gen, inclusionAI (Ant Group) · ~41 GB VRAM · commercial OK
10. CSM-1B (Sesame) — Text → speech, Sesame · ~8.0 GB VRAM · commercial OK
11. StyleTTS 2 — Text → speech, yl4579 · ~2.0 GB VRAM · non-commercial
12. Parler-TTS Large v1 — Text → speech, Parler-TTS (HF/Descript) · ~9.0 GB VRAM · commercial OK
13. MeloTTS — Text → speech, MyShell.ai / MIT · ~1.0 GB VRAM · commercial OK

More: all "best" lists · cost calculator · all models