import gradio as gr
import librosa
import numpy as np
import torch

from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan

checkpoint = "burraco135/speecht5_finetuned_voxpopuli_it"
processor = SpeechT5Processor.from_pretrained(checkpoint)
model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

speaker_embeddings = r"C:\Users\ester\OneDrive\Documenti\VSCode\Hugging Face\tts\speaker_embeddings.npy"

def predict(text, speaker):
    # se il testo è vuoto, fai un array vuoto da 16000
    if len(text.strip()) == 0:
        return (16000, np.zeros(0).astype(np.int16))

    # preprocess text
    inputs = processor(text=text, return_tensors="pt")

    vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
    speech = model.generate_speech(speaker_embeddings, vocoder=vocoder)
    return (16000, speech)


title = "SpeechT5: Speech Synthesis"

examples = [
    ["It is not in the stars to hold our destiny but in ourselves.", "BDL (male)"],
    ["The octopus and Oliver went to the opera in October.", "CLB (female)"],
    ["She sells seashells by the seashore. I saw a kitten eating chicken in the kitchen.", "RMS (male)"],
    ["Brisk brave brigadiers brandished broad bright blades, blunderbusses, and bludgeons—balancing them badly.", "SLT (female)"],
    ["A synonym for cinnamon is a cinnamon synonym.", "BDL (male)"],
    ["How much wood would a woodchuck chuck if a woodchuck could chuck wood? He would chuck, he would, as much as he could, and chuck as much wood as a woodchuck would if a woodchuck could chuck wood.", "CLB (female)"],
]

gr.Interface(
    fn=predict,
    inputs=[
        gr.Text(label="Input Text")
    ],
    outputs=[
        gr.Audio(label="Generated Speech", type="numpy"),
    ],
    title=title,
    examples=examples,
).launch()