VibeVoice Egyptian Arabic — cfg_scale=5.0
Fine-tuned VibeVoice on Egyptian Arabic dialect (checkpoint-9160).
cfg_scale=5.0
Strongest voice cloning fidelity. More compact generation.
Higher guidance = shorter output; verify audio covers all script lines.
Repo contents
| File | Description |
|---|---|
model.safetensors |
Merged model weights (5.1 GB, single shard) |
config.json |
Model architecture config |
tokenizer.json + friends |
Qwen2.5 tokenizer files |
preprocessor_config.json |
Audio processor settings |
voices/egyptian_male.wav |
Reference voice for male speaker |
voices/egyptian_female.wav |
Reference voice for female speaker |
samples/demo_cfg5.0.wav |
Sample output at this cfg_scale |
Backend usage
from huggingface_hub import hf_hub_download, snapshot_download
snapshot_download("MohammedEhab20/vibe-voice-egyptian-cfg50", local_dir="./model")
male_voice = "./model/voices/egyptian_male.wav"
female_voice = "./model/voices/egyptian_female.wav"
# Run inference:
# python inference_from_file.py --model_path ./model --cfg_scale 5.0 ...
- Downloads last month
- 6
import torch, soundfile as sf, librosa, numpy as np from vibevoice.processor.vibevoice_processor import VibeVoiceProcessor from vibevoice.modular.modeling_vibevoice_inference import VibeVoiceForConditionalGenerationInference # Load voice sample (should be 24kHz mono) voice, sr = sf.read("path/to/voice_sample.wav") if voice.ndim > 1: voice = voice.mean(axis=1) if sr != 24000: voice = librosa.resample(voice, sr, 24000) processor = VibeVoiceProcessor.from_pretrained("MohammedEhab20/vibe-voice-egyptian-cfg50") model = VibeVoiceForConditionalGenerationInference.from_pretrained( "MohammedEhab20/vibe-voice-egyptian-cfg50", torch_dtype=torch.bfloat16 ).to("cuda").eval() model.set_ddpm_inference_steps(5) inputs = processor(text=["Speaker 0: Hello!\nSpeaker 1: Hi there!"], voice_samples=[[voice]], return_tensors="pt") audio = model.generate(**inputs, cfg_scale=1.3, tokenizer=processor.tokenizer).speech_outputs[0] sf.write("output.wav", audio.cpu().numpy().squeeze(), 24000)