Adding Voice preset in AutoProcessor Giving error

[HarshPatil89]

Below is my code Its very simple.

`from transformers import AutoProcessor, BarkModel

import scipy

import torch

processor = AutoProcessor.from_pretrained("suno/bark")
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = BarkModel.from_pretrained("suno/bark").to(device)
voice_preset = "v2/en_speaker_4"
inputs = processor("Hello, my dog is cute, I need him in my life", voice_preset=voice_preset).to(device)
audio_array = model.generate(**inputs)
audio_array = audio_array.cpu().numpy().squeeze()
sample_rate = 24000
scipy.io.wavfile.write("bark_out2.wav", rate=sample_rate, data=audio_array)`

When I add voice_preset as an argument it gives error regarding

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)

[gruellan]

inputs = processor("Hello, my dog is cute, I need him in my life", voice_preset=voice_preset).to(device) returns a dictionary, not a tensor. Calling .to(device) on the whole dict won't automatically put the tensors inside inputs on the GPU.
You should instead iterate through and put each tensor onto GPU:

inputs = processor("Hello, my dog is cute, I need him in my life", voice_preset=voice_preset)
for key, value in inputs.items():
    inputs[key] = inputs[key].to(device)

audio_array = model.generate(**inputs)

Tested this change to your code and it worked