app.py

###################################### imports ######################################
import torch
from TTS.api import TTS
import gradio as gr
import os
import spaces
import yaml


###################################### utilities ######################################
def get_config():
    # get config path
    config_path = os.environ["CONFIG_PATH"]
    # Parse the YAML file
    with open(config_path, 'r') as file:
        config = yaml.safe_load(file)
    
    return config


def init_TTS(config):
    # Get device
    device = "cuda" if torch.cuda.is_available() else "cpu"
    # Initialize the TTS model
    tts = TTS(config['inference']['model']).to(device)
    return tts


@spaces.GPU
def generate_speech(voice_choice, markdown, microphone, text):
    # Generate speech using the provided text, speaker voice, and language

    if voice_choice=="Record":
        speaker = microphone
    
    else:
        speaker = config['inference']['speaker_wav']
    
    tts.tts_to_file(text=text,
                    file_path=config['inference']['file_path'],
                    speaker_wav=speaker,
                    language=config['inference']['language'])
    return config['inference']['file_path']


###################################### main ######################################
def UI(config):

    # gradio elements
    voice_choice = gr.Radio(label="Record or use a predefined voice.", 
                            choices=["Record", "Predefined (Nancy)"], 
                            value="Record")
    markdown = gr.Markdown("""If recording, speak loud and clearly. Recommended speaking track 
                              '*printing, in the only sense with which we are at present concerned, 
                              differs from, most if not all, the arts and crafts in the exhibition.*'""")
    microphone = gr.Audio(label="Audio", sources="microphone", type="filepath", elem_id='audio')
    enter_text = gr.Textbox(label="Enter your text")
    
    # Create the Gradio interface
    demo = gr.Interface(
        fn=generate_speech,
        inputs=[
            voice_choice,
            markdown,
            microphone,
            enter_text
        ],
        outputs="audio",
        title="Voice cloning and Synthesis with Coqui-XTTS",
        description="Clone your voice and Synthesize speech using predefined target voice and language. It takes a 10-20 seconds to download the model, so wait to record until the app is *Running on Zero* to begin."
    )

    # Launch the interface
    demo.launch()
    return 0


###################################### Execute ######################################
if __name__ == "__main__":
    # Get config
    config = get_config()
    
    # initialize TTS
    tts = init_TTS(config)
    
    # run program
    UI(config)