ai-agents-device-mismatch-when-using-barkprocessor-with-a-voice-preset-and-g-58a0a6ea

THEN Manually load the voice_preset from the npz file, create a dictionary with keys 'fine_prompt', 'coarse_prompt', 'semantic_prompt', then before calling inputs.to(device), explicitly move inputs['history_prompt'] to the target device with .to(device). This ensures all tensors are on the same device before generation.

import numpy as np
with np.load('bark/assets/prompts/en_speaker_6.npz') as data:
    voice_preset = {
        'fine_prompt': data['fine_prompt'],
        'coarse_prompt': data['coarse_prompt'],
        'semantic_prompt': data['semantic_prompt']
    }
inputs = processor(text_prompt, voice_preset=voice_preset)
inputs['history_prompt'] = inputs['history_prompt'].to(device)
inputs = inputs.to(device)