Best local text-to-speech models

Speech synthesis models you can run yourself. Ranking these alongside transcription models, as a single 'speech' list would, compares two unrelated tasks.

From the file· live filter over real dataFrom the file· 23 models

How this is ranked

Ranked by downloads among synthesis models. Streaming latency matters more than throughput here, and nobody has measured it on consumer hardware — including us.

Best local text-to-speech models

#	Model	Params○	Smallest quant●	Smallest●
1	Ace-Step1.5	160M	0.04 GiB	0.04 GiB
2	Qwen3-TTS-12Hz-0.6B-Base	915M	0.50 GiB	0.50 GiB
3	OmniVoice	613M	0.56 GiB	0.56 GiB
4	orpheus-3b-0.1-ft	3.8B	0.91 GiB	0.91 GiB
5	VieNeu-TTS-0.3B	244M	0.19 GiB	0.19 GiB
6	neutts-air	748M	0.43 GiB	0.43 GiB
7	s2-proMoE	4.6B	2.40 GiB	2.40 GiB
8	Fun-CosyVoice3-0.5B-2512	—	0.34 GiB	0.34 GiB
9	VoxCPM2	2.3B	1.57 GiB	1.57 GiB
10	chatterbox	264M	0.17 GiB	0.17 GiB
11	Kokoro-82M	82M	0.13 GiB	0.13 GiB
12	VieNeu-TTS	553M	0.39 GiB	0.39 GiB
13	csm-1b	1.6B	1.34 GiB	1.34 GiB
14	VibeVoice-Realtime-0.5B	1.0B	0.65 GiB	0.65 GiB
15	tada-3b-ml	4.2B	5.20 GiB	5.20 GiB
16	MOSS-TTS-Local-Transformer-v1.5	4.6B	5.56 GiB	5.56 GiB
17	Qwen3-TTS-12Hz-1.7B-Base	1.9B	0.94 GiB	0.94 GiB
18	orpheus-3b-0.1-pretrained	3.8B	1.40 GiB	1.40 GiB
19	ZONOS2	—	4.58 GiB	4.58 GiB
20	Qwen3-TTS-12Hz-1.7B-VoiceDesign	1.9B	1.90 GiB	1.90 GiB
21	VibeVoice-1.5B	2.7B	1.76 GiB	1.76 GiB
22	VoxCPM-0.5B	728M	2.45 GiB	2.45 GiB
23	Qwen3-TTS-12Hz-1.7B-CustomVoice	1.9B	1.90 GiB	1.90 GiB

○Spec sheet●From the file●From the filewhat these mean

Best local models for coding Best local reasoning models Best models for an 8GB card Best models for long context Best permissively licensed models Best models under 4GB Largest models with a runnable quantization Best local speech recognition models Best local vision-language models Best local image generation models