app.py

import os
from huggingface_hub import hf_hub_download
import gradio as gr
from piper import PiperVoice
from io import BytesIO
import wave
import numpy as np


def text_to_speech(text):
    # Load voice data
    model_path = hf_hub_download(repo_id="sekhan/luxembourgish-voice",
                                    repo_type='dataset',
                                    filename="high/lu_rtl_high3239.onnx",
                                    token=os.environ['HF_TOKEN'])
    config_path = hf_hub_download(repo_id="sekhan/luxembourgish-voice",
                                    repo_type='dataset',
                                    filename="high/lu_rtl_high3239.onnx.json",
                                    token=os.environ['HF_TOKEN'])

    # Load Lux. voice
    voice = PiperVoice.load(model_path, config_path)

    buffer = BytesIO()
    with wave.open(buffer, 'wb') as wav_file:
        wav_file.setframerate(voice.config.sample_rate)
        wav_file.setsampwidth(2)
        wav_file.setnchannels(1)
        voice.synthesize(text, wav_file, sentence_silence=0.5, length_scale=1.1, noise_scale=0.75)
    buffer.seek(0)
    audio_data = np.frombuffer(buffer.read(), dtype=np.int16)

    return audio_data.tobytes(), None


# Gradio Interface
with gr.Blocks(theme=gr.themes.Base(), css="footer {visibility: hidden}") as blocks:
    gr.Markdown("# Luxembourgish Text-to-Speech Synthesizer")
    gr.Markdown("Enter Luxembourgish text to synthesize it into speech. This is a very early demo. Your spontaneous text data are not saved and only used for the speech synthesis.")
    input_text = gr.Textbox(label="Input Text", max_lines=3, placeholder="Enter text here...")
    submit_button = gr.Button("Synthesize")
    output_audio = gr.Audio(label="Synthesized Speech", type="numpy", show_download_button=False)
    output_text = gr.Textbox(label="Output Text", visible=False)

    def process_and_output(text):
        audio, message = text_to_speech(text)
        if message:
            return audio, message
        else:
            return audio, None

    submit_button.click(process_and_output, inputs=input_text, outputs=[output_audio, output_text])

blocks.launch()