import gradio as gr
import wave
import numpy as np
from io import BytesIO
from huggingface_hub import hf_hub_download
from piper import PiperVoice
from transformers import pipeline
import typing
import random

model_path = hf_hub_download(repo_id="larcanio/piper-voices", filename="es_AR-daniela-high.onnx")
config_path = hf_hub_download(repo_id="larcanio/piper-voices", filename="es_AR-daniela-high.json")
voice = PiperVoice.load(model_path, config_path)
with open('assets/sentences_es.txt', 'r') as r:
    random_quotes = [line.strip() for line in r]

def synthesize_speech(text):
    # Create an in-memory buffer for the WAV file
    buffer = BytesIO()
    with wave.open(buffer, 'wb') as wav_file:
        wav_file.setframerate(voice.config.sample_rate)
        wav_file.setsampwidth(2)  # 16-bit
        wav_file.setnchannels(1)  # mono

        # Synthesize speech
        # eztext = preprocess_text(text)
        voice.synthesize(text, wav_file)

    # Convert buffer to NumPy array for Gradio output
    buffer.seek(0)
    audio_data = np.frombuffer(buffer.read(), dtype=np.int16)

    return audio_data.tobytes(), None

def get_random_quote():
    return random.choice(random_quotes)

BANNER_TEXT = """
# Demo en español argentino con Piper

[***Piper***](https://huggingface.co/rhasspy/piper-voices/) es un modelo de abierto de Texto a Voz (TTS)
que permite entrenarse con voz propia, destaca por no requerir conectarse a Internet y ofrecer resultados
sin exigir GPU.  Inicialmente diseñado para Raspberri Pi.

Este demo solo muestra español, puedes probar [voces en otros idiomas](https://rhasspy.github.io/piper-samples/).
"""

FOOTER_TEXT = """
# Credits

[voice trained](https://huggingface.co/larcanio/piper-voices) by [larcanio](https://huggingface.co/larcanio/),
[original demo](https://huggingface.co/gyroing/Persian-Piper-Model-gyro) by [gyroing](https://huggingface.co/gyroing/)
on [piper's shoulders](https://huggingface.co/rhasspy/piper-voices) by [rhasspy](https://github.com/rhasspy). [More info](https://huggingface.co/spaces/igortamara/sample-tts-piper/blob/main/README.md)
"""

# Using Gradio Blocks
with gr.Blocks(theme=gr.themes.Base(), title="Piper Argentinian voice test") as demo:
    gr.Markdown(BANNER_TEXT)
    input_text = gr.Textbox(label=" ", placeholder="Introduce el texto a leer aquí")
    with gr.Row():
        submit_button = gr.Button("Genera audio")
        random_btn = gr.Button('🎲 Cita aleatoria 💬', variant='secondary')
    output_audio = gr.Audio(label="Audio generado", type="numpy", interactive=False, streaming=False, autoplay=True)
    output_text = gr.Textbox(label="Tokens generados", visible=False)

    gr.Markdown(FOOTER_TEXT)

    submit_button.click(synthesize_speech, inputs=input_text, outputs=[output_audio, output_text])
    random_btn.click(fn=get_random_quote, inputs=[], outputs=[input_text])


if __name__ == '__main__':
    demo.launch()