TTS Bench — Speed — linux-cloning

Rig: linux-3090 — AMD Ryzen 9 5900XT 16-Core Processor · NVIDIA GeForce RTX 3090 24GB · 63 GB RAM · Linux 6.8.0-117-generic

Label: cloning · chris_hemsworth_15s — ref chris_hemsworth_15s.wav

TTFA = time to first audio (ms; lower is better). RTF = real-time factor (× realtime; higher is better; e.g. 10× means 10 sec of audio generated per 1 sec of compute). Cold = first run after process start; warm = subsequent runs.

Speed winners

Fastest (this cloning run): OmniVoice (cuda) — 4.99× warm RTF, 1.49s warm TTFA

Model	Device	TTFA cold	TTFA warm	RTF cold	RTF warm	Peak RAM	Peak VRAM	Size
LuxTTS	cuda	436ms	241ms	18.76×	34.86×	2.35 GB	1.13 GB	—
OmniVoice	cuda	2.01s	1.49s	3.80×	4.99×	2.05 GB	2.33 GB	~1B
Coqui XTTS-v2	cuda	1.90s	1.53s	3.63×	4.77×	2.26 GB	2.08 GB	750M
Chatterbox Turbo	cuda	2.12s	1.61s	3.10×	4.28×	2.34 GB	3.28 GB	744M
Pocket-TTS	cpu	151ms	151ms	3.94×	3.86×	1.69 GB	—	100M
F5-TTS	cuda	2.49s	1.84s	2.33×	3.17×	2.61 GB	882 MB	330M
Qwen3-TTS 1.7B (CUDA-graph)	cuda	51.60s	19.43s	2.86×	3.07×	2.33 GB	4.95 GB	1.7B
NeuTTS Nano	cuda	952ms	357ms	1.85×	2.64×	3.45 GB	3.25 GB	748M
LuxTTS	cpu	2.92s	2.84s	2.14×	2.20×	2.62 GB	—	—
Chatterbox	cuda	3.90s	3.11s	1.71×	2.16×	2.20 GB	3.66 GB	1.2B
MOSS-TTS-Nano	cuda	4.61s	3.22s	1.54×	2.14×	1.91 GB	870 MB	100M
VoxCPM2 2B	cuda	4.24s	3.10s	1.44×	2.05×	3.37 GB	5.82 GB	2B
NeuTTS Nano	cpu	1.01s	419ms	1.47×	1.68×	5.55 GB	—	748M
NeuTTS Air	cuda	1.64s	478ms	1.16×	1.36×	3.88 GB	3.25 GB	748M
NeuTTS Air	cpu	1.68s	587ms	1.07×	1.33×	5.98 GB	—	748M
IndexTTS-2	cpu	6.72s	5.86s	0.97×	1.15×	2.52 GB	—	1.5B
MOSS-TTS-Nano	cpu	6.85s	6.63s	1.00×	1.11×	1.31 GB	—	100M
IndexTTS-2	cuda	7.72s	6.49s	0.90×	1.07×	2.80 GB	7.70 GB	1.5B
Coqui XTTS-v2	cpu	11.08s	10.05s	0.77×	0.79×	3.09 GB	—	750M
Sesame CSM-1B	cuda	9.14s	8.23s	0.62×	0.74×	2.24 GB	3.54 GB	1B
Chatterbox Turbo	cpu	10.02s	10.36s	0.63×	0.67×	4.08 GB	—	744M
Dia 1.6B	cuda	12.83s	9.94s	0.42×	0.54×	2.11 GB	4.66 GB	1.6B
ZipVoice 123M (4/5 ok)	cpu	15.97s	12.22s	0.41×	0.52×	51.51 GB	—	123M
Chatterbox	cpu	21.31s	20.76s	0.32×	0.33×	4.50 GB	—	1.2B
VoxCPM2 2B	cpu	34.02s	33.01s	0.17×	0.18×	6.54 GB	—	2B
OmniVoice	cpu	49.48s	47.44s	0.15×	0.16×	4.12 GB	—	~1B
Sesame CSM-1B	cpu	50.29s	51.41s	0.11×	0.12×	6.38 GB	—	1B
F5-TTS	cpu	75.56s	76.09s	0.08×	0.08×	2.63 GB	—	330M
Mars5-TTS	cuda	87.24s	87.80s	0.05×	0.05×	2.60 GB	7.24 GB	1.2B
Mars5-TTS	cpu	93.19s	85.83s	0.06×	0.05×	2.59 GB	—	1.2B
ZipVoice 123M	cuda	Skipped — out of GPU memory (model exceeds this GPU's VRAM)

Model

Device

TTFA cold

TTFA warm

RTF cold

RTF warm

Peak RAM

Peak VRAM

Size

LuxTTS

cuda

436ms

241ms

18.76×

34.86×

2.35 GB

1.13 GB

—

OmniVoice

cuda

2.01s

1.49s

3.80×

4.99×

2.05 GB

2.33 GB

~1B

Coqui XTTS-v2

cuda

1.90s

1.53s

3.63×

4.77×

2.26 GB

2.08 GB

750M

Chatterbox Turbo

cuda

2.12s

1.61s

3.10×

4.28×

2.34 GB

3.28 GB

744M

Pocket-TTS

cpu

151ms

3.94×

3.86×

1.69 GB

—

100M

F5-TTS

cuda

2.49s

1.84s

2.33×

3.17×

2.61 GB

882 MB

330M

Qwen3-TTS 1.7B (CUDA-graph)

cuda

51.60s

19.43s

2.86×

3.07×

2.33 GB

4.95 GB

1.7B

NeuTTS Nano

cuda

952ms

357ms

1.85×

2.64×

3.45 GB

3.25 GB

748M

LuxTTS

cpu

2.92s

2.84s

2.14×

2.20×

2.62 GB

—

Chatterbox

cuda

3.90s

3.11s

1.71×

2.16×

2.20 GB

3.66 GB

1.2B

MOSS-TTS-Nano

cuda

4.61s

3.22s

1.54×

2.14×

1.91 GB

870 MB

100M

VoxCPM2 2B

cuda

4.24s

3.10s

1.44×

2.05×

3.37 GB

5.82 GB

NeuTTS Nano

cpu

1.01s

419ms

1.47×

1.68×

5.55 GB

—

748M

NeuTTS Air

cuda

1.64s

478ms

1.16×

1.36×

3.88 GB

3.25 GB

748M

NeuTTS Air

cpu

1.68s

587ms

1.07×

1.33×

5.98 GB

—

748M

IndexTTS-2

cpu

6.72s

5.86s

0.97×

1.15×

2.52 GB

—

1.5B

MOSS-TTS-Nano

cpu

6.85s

6.63s

1.00×

1.11×

1.31 GB

—

100M

IndexTTS-2

cuda

7.72s

6.49s

0.90×

1.07×

2.80 GB

7.70 GB

1.5B

Coqui XTTS-v2

cpu

11.08s

10.05s

0.77×

0.79×

3.09 GB

—

750M

Sesame CSM-1B

cuda

9.14s

8.23s

0.62×

0.74×

2.24 GB

3.54 GB

Chatterbox Turbo

cpu

10.02s

10.36s

0.63×

0.67×

4.08 GB

—

744M

Dia 1.6B

cuda

12.83s

9.94s

0.42×

0.54×

2.11 GB

4.66 GB

1.6B

ZipVoice 123M (4/5 ok)

cpu

15.97s

12.22s

0.41×

0.52×

51.51 GB

—

123M

Chatterbox

cpu

21.31s

20.76s

0.32×

0.33×

4.50 GB

—

1.2B

VoxCPM2 2B

cpu

34.02s

33.01s

0.17×

0.18×

6.54 GB

—

OmniVoice

cpu

49.48s

47.44s

0.15×

0.16×

4.12 GB

—

~1B

Sesame CSM-1B

cpu

50.29s

51.41s

0.11×

0.12×

6.38 GB

—

F5-TTS

cpu

75.56s

76.09s

0.08×

2.63 GB

—

330M

Mars5-TTS

cuda

87.24s

87.80s

0.05×

2.60 GB

7.24 GB

1.2B

Mars5-TTS

cpu

93.19s

85.83s

0.06×

0.05×

2.59 GB

—

1.2B

ZipVoice 123M

cuda

Skipped — out of GPU memory (model exceeds this GPU's VRAM)