AImodelsComparison.com 1,011 models
Menu
Models Best of Trending Can I run it? Methodology About
1,011 models · refreshed nightly

Text to speech models

Every model in the catalog with its licence, estimated VRAM and daily-tracked downloads. Filters update the URL — share any view.

Task
All text-generation 400 text-to-image 150 ASR 89 image-text-to-text 82 sentence-similarity 55 feature-extraction 47 translation 44 text-classification 39 image-classification 32 token-classification 22 text-to-speech 16 audio-classification 12 image-segmentation 8 object-detection 8 summarization 7
Runs on
Any RTX 3060 · 12 GB RTX 4070 · 16 GB RTX 4090 · 24 GB 2 × 24 GB Mac M3 · 24 GB Mac M4 Max · 64 GB
Licence
Size
$
16 models
#ModelParamsContextCommercial use30dMin VRAM
01 Kokoro-82M hexgrad · text-to-speech — — ✓ apache-2.0 12.0 M —
02 XTTS-v2 coqui · text-to-speech — — other 9.2 M —
03 chatterbox ResembleAI · text-to-speech — — ✓ mit 2.5 M from 12.2 GB
04 Qwen3-TTS-12Hz-1.7B-CustomVoice Qwen · text-to-speech 1.9 B — ✓ apache-2.0 2.3 M from 5.8 GB
05 Qwen3-TTS-12Hz-0.6B-CustomVoice Qwen · text-to-speech 910 M — ✓ apache-2.0 1.6 M from 3.4 GB
06 Kokoro-82M-v1.0-ONNX onnx-community · text-to-speech — — ✓ apache-2.0 1.4 M —
07 VoxCPM2 openbmb · text-to-speech 2.3 B — ✓ apache-2.0 900 K from 5.9 GB
08 OmniVoice k2-fsa · text-to-speech 610 M — unknown 877 K from 4.2 GB
09 F5-TTS SWivid · text-to-speech — — ✗ cc-by-nc-4.0 770 K from 5.0 GB
10 VibeVoice-Realtime-0.5B microsoft · text-to-speech 1.0 B — ✓ mit 657 K from 2.9 GB
11 Qwen3-TTS-12Hz-1.7B-VoiceDesign Qwen · text-to-speech 1.9 B — ✓ apache-2.0 574 K from 5.8 GB
12 Qwen3-TTS-12Hz-0.6B-Base Qwen · text-to-speech 910 M — ✓ apache-2.0 445 K from 3.4 GB
13 MOSS-TTS OpenMOSS-Team · text-to-speech 8.5 B — ✓ apache-2.0 375 K from 20.5 GB
14 higgs-tts-3-4b bosonai · text-to-speech 4.7 B — other 374 K from 11.4 GB
15 VieNeu-TTS-v3-Turbo pnnbao-ump · text-to-speech 130 M 1 K ✓ apache-2.0 370 K from 1.1 GB
16 s2-pro fishaudio · text-to-speech 4.6 B — other 368 K from 11.2 GB
VRAM figures are estimates for the smallest available quantization at 8K context — see /methodology. Downloads refresh nightly from the Hugging Face API.