app.py

import tempfile

import edge_tts
import gradio as gr
from transformers import pipeline
import pyarabic.araby as araby

language_dict = {
  "English": {
    "Jenny": "en-US-JennyNeural",
    "Guy": "en-US-GuyNeural",
    "Ana": "en-US-AnaNeural",
    "Aria": "en-US-AriaNeural",
    "Christopher": "en-US-ChristopherNeural",
    "Eric": "en-US-EricNeural",
    "Michelle": "en-US-MichelleNeural",
    "Roger": "en-US-RogerNeural",
    "Natasha": "en-AU-NatashaNeural",
    "William": "en-AU-WilliamNeural",
    "Clara": "en-CA-ClaraNeural",
    "Liam": "en-CA-LiamNeural",
    "Libby": "en-GB-LibbyNeural",
    "Maisie": "en-GB-MaisieNeural",
    "Ryan": "en-GB-RyanNeural",
    "Sonia": "en-GB-SoniaNeural",
    "Thomas": "en-GB-ThomasNeural",
    "Sam": "en-HK-SamNeural",
    "Yan": "en-HK-YanNeural",
    "Connor": "en-IE-ConnorNeural",
    "Emily": "en-IE-EmilyNeural",
    "Neerja": "en-IN-NeerjaNeural",
    "Prabhat": "en-IN-PrabhatNeural",
    "Asilia": "en-KE-AsiliaNeural",
    "Chilemba": "en-KE-ChilembaNeural",
    "Abeo": "en-NG-AbeoNeural",
    "Ezinne": "en-NG-EzinneNeural",
    "Mitchell": "en-NZ-MitchellNeural",
    "James": "en-PH-JamesNeural",
    "Rosa": "en-PH-RosaNeural",
    "Luna": "en-SG-LunaNeural",
    "Wayne": "en-SG-WayneNeural",
    "Elimu": "en-TZ-ElimuNeural",
    "Imani": "en-TZ-ImaniNeural",
    "Leah": "en-ZA-LeahNeural",
    "Luke": "en-ZA-LukeNeural"
  },
  "Vietnamese": {
    "HoaiMy": "vi-VN-HoaiMyNeural",
    "NamMinh": "vi-VN-NamMinhNeural"
  }
}

pipe = pipeline("text2text-generation", model="mush42/fine-tashkeel")

async def text_to_speech_edge(text, language_code, speaker, tashkeel_checkbox=False, speed=0, pitch=0):

  # Remove diacritics from Arabic text then add tashkeel
    if language_code == "Arabic" and tashkeel_checkbox:
      text = pipe(araby.strip_diacritics(text))[0]["generated_text"].strip()
    
    # Get the voice for the selected language and speaker
    voice = language_dict[language_code][speaker]
    rate_str = f"{speed:+d}%"
    pitch_str = f"{pitch:+d}Hz"
    communicate = edge_tts.Communicate(text, voice, rate=rate_str, pitch=pitch_str)
    
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
      tmp_path = tmp_file.name
      await communicate.save(tmp_path)

    return text, tmp_path


def get_speakers(language):
    print(language)
    speakers = list(language_dict[language].keys())
    return gr.Dropdown(choices=speakers, value=speakers[0], interactive=True), gr.Checkbox(visible=language == "Arabic", interactive=True)


default_language = None
default_speaker = None
with gr.Blocks(title="Inspire4DEV - Đọc Văn Bản") as demo:
    gr.HTML("<center><h1>CHUYỂN VĂN BẢN THÀNH GIỌNG ĐỌC</h1></center>")
    gr.HTML(f"<h2 style='color:Tomato;'> {len(language_dict)} ngôn ngữ được hỗ trợ {', '.join(language_dict.keys())}</h3>")
    gr.HTML(f"<p> {', '.join(language_dict.keys())} </h3>")
    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(lines=5, label="Văn bản", placeholder="Hãy nhập văn bản cần chuyển thành giọng đọc")
            language = gr.Dropdown(
                choices=list(language_dict.keys()), value=default_language, label="Lựa chọn ngôn ngữ", interactive=True
            )
            speaker = gr.Dropdown(choices=[], value=default_speaker, label="Giọng Đọc", interactive=False)
            tashkeel_checkbox = gr.Checkbox(label="Tashkeel", value=False, visible=False, interactive=False)
            speedrate = gr.Slider(minimum=-50, maximum=50, value=0, label="Tốc độ đọc (%)", step=1)
            pitchadj = gr.Slider(minimum=-20, maximum=20, value=0, label="Tông giọng (Hz)", step=1)
            run_btn = gr.Button(value="TẠO GIỌNG ĐỌC", variant="primary")

        with gr.Column():
            output_text = gr.Textbox(label="Văn bản sau khi chuyển đổi")
            output_audio = gr.Audio(type="filepath", label="Âm thanh")

    language.change(get_speakers, inputs=[language], outputs=[speaker, tashkeel_checkbox])
    run_btn.click(text_to_speech_edge, inputs=[input_text, language, speaker, tashkeel_checkbox, speedrate, pitchadj], outputs=[output_text, output_audio])

if __name__ == "__main__":
    demo.queue().launch(share=False)