Views
No views yet
1from vibevoice.modular.modeling_vibevoice_inference import VibeVoiceForConditionalGenerationInference
2from vibevoice.processor.vibevoice_processor import VibeVoiceProcessor
3
4model = VibeVoiceForConditionalGenerationInference.from_pretrained(
5 "Dannidee/VibeVoice7b-low-vram/4bit",
6 device_map='cuda',
7 torch_dtype=torch.bfloat16,
8)
9processor = VibeVoiceProcessor.from_pretrained("Dannidee/VibeVoice7b-low-vram/4bit")
10
11# Generate speech
12text = "Speaker 1: Hello! Speaker 2: Hi there!"
13inputs = processor(
14 text=[text],
15 voice_samples=[["voice1.wav", "voice2.wav"]],
16 padding=True,
17 return_tensors="pt",
18)
19
20outputs = model.generate(**inputs)
21processor.save_audio(outputs.speech_outputs[0], "output.wav")