from transformers import VitsModel, AutoTokenizer
import torch
import scipy.io.wavfile as wavfile

model = VitsModel.from_pretrained("spanish-tts")
tokenizer = AutoTokenizer.from_pretrained("spanish-tts")
text=    "Estoy llamando desde la oficina de impuestos."

inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    output = model(**inputs).waveform  # Output is a tensor of shape [batch_size, waveform_length]

output_audio = output.squeeze(0).numpy()  # Remove batch dimension
wavfile.write("spanish.wav", rate=model.config.sampling_rate, data=output_audio)