VoxCPM-0.5B — GGUF weights for llama.cpp-omni
GGUF-converted weights of openbmb/VoxCPM-0.5B
for the C++/ggml engine llama.cpp-omni
(tools/omni/voxcpm2). The lightest VoxCPM — runs TTS + voice cloning natively on
CPU / Metal / CUDA / Vulkan, no PyTorch runtime required.
Files
| File | Format | Size | Component |
|---|---|---|---|
VoxCPM-0.5B-BaseLM-F16.gguf |
F16 | ~974 MB | Base language model |
VoxCPM-0.5B-BaseLM-Q8_0.gguf |
Q8_0 | ~519 MB | Base language model, 8-bit (recommended) |
VoxCPM-0.5B-Acoustic-F16.gguf |
F16 | ~560 MB | Acoustic stack (ResidualLM + FSQ + CFM + AudioVAE) |
Output: 16 kHz mono. Use one BaseLM (Q8_0 recommended) + the Acoustic file.
Usage
./voxcpm2-cli -t "Hello from VoxCPM 0.5B." -o out.wav \
VoxCPM-0.5B-BaseLM-Q8_0.gguf VoxCPM-0.5B-Acoustic-F16.gguf
# voice cloning: add -r speaker.wav
Verified on Apple M4 Pro / Metal (RTF ~1.06 — fastest of the VoxCPM family).
Flags: --cfg, --timesteps, --seed, --temperature, -r (clone),
--prompt-wav/--prompt-text (reference-transcript clone), --cpu.
License/terms inherit from the upstream model.
- Downloads last month
- 257
Hardware compatibility
Log In to add your hardware
8-bit
16-bit
import soundfile as sf from voxcpm import VoxCPM model = VoxCPM.from_pretrained("DennisHuang648/VoxCPM-0.5B-GGUF") wav = model.generate( text="VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.", prompt_wav_path=None, # optional: path to a prompt speech for voice cloning prompt_text=None, # optional: reference text cfg_value=2.0, # LM guidance on LocDiT, higher for better adherence to the prompt, but maybe worse inference_timesteps=10, # LocDiT inference timesteps, higher for better result, lower for fast speed normalize=True, # enable external TN tool denoise=True, # enable external Denoise tool retry_badcase=True, # enable retrying mode for some bad cases (unstoppable) retry_badcase_max_times=3, # maximum retrying times retry_badcase_ratio_threshold=6.0, # maximum length restriction for bad case detection (simple but effective), it could be adjusted for slow pace speech ) sf.write("output.wav", wav, 16000) print("saved: output.wav")