app.py

import logging
from typing import cast
from pdfminer.high_level import extract_text
from huggingface_hub import hf_hub_download, list_repo_files
import gradio as gr
from balacoon_tts import TTS

# Global TTS module, initialized from a selected model
tts = None
speakers = []
default_speaker = None

def read_pdf(file):
    text = extract_text(file.name)
    return text

def synthesize_audio(file, speaker_str):
    text_str = read_pdf(file)
    if len(text_str) > 1024:
        text_str = text_str[:1024]
    samples = tts.synthesize(text_str, speaker_str)
    return (samples, tts.get_sampling_rate())

def main():
    logging.basicConfig(level=logging.INFO)
    file_input = gr.inputs.File(label="Upload PDF")

    model_files = list_repo_files(repo_id="balacoon/tts")
    model_name_dropdown = gr.inputs.Dropdown(label="Model", choices=model_files)

    def set_model(model_name_str: str):
        model_path = hf_hub_download(repo_id="balacoon/tts", filename=model_name_str)
        global tts, speakers, default_speaker
        tts = TTS(model_path)
        speakers = tts.get_speakers()
        default_speaker = speakers[-1]

    speaker_dropdown = gr.inputs.Dropdown(label="Speaker", choices=speakers)

    audio = gr.outputs.Audio(label="Generated Audio", type="numpy")

    def generate_audio(file, model_name_str, speaker_str):
        set_model(model_name_str)
        return synthesize_audio(file, speaker_str)

    iface = gr.Interface(
        fn=generate_audio,
        inputs=[file_input, model_name_dropdown, speaker_dropdown],
        outputs=audio,
        title="PDF TO SPEECH CONVERTER",
        layout="rows",
        debug=True
    )

    iface.launch()

if __name__ == "__main__":
    main()