app.py

# app.py - fixed version for dialog support with edge-tts SSML
import os
import tempfile
import edge_tts
import gradio as gr
from xml.sax.saxutils import escape

def build_ssml_from_dialog(text: str, default_voice: str = "fa-IR-DilaraNeural") -> str:
    """
    Parses input line by line, detects 'مرد:' or 'زن:' prefixes,
    and wraps each segment in <voice> tags with a short break.
    """
    lines = [l.strip() for l in text.splitlines() if l.strip()]
    if not lines:
        return ""

    voice_map = {
        "مرد:": "fa-IR-FaridNeural",
        "زن:": "fa-IR-DilaraNeural",
    }

    segments = []
    for line in lines:
        chosen_voice = default_voice
        content = line
        for prefix, voice_name in voice_map.items():
            if line.startswith(prefix):
                content = line[len(prefix):].strip()
                chosen_voice = voice_name
                break
        content_escaped = escape(content)
        segments.append(f'<voice name="{chosen_voice}">{content_escaped}<break time="300ms"/></voice>')

    ssml = '<?xml version="1.0" encoding="utf-8"?><speak version="1.0" xml:lang="fa-IR">' + "".join(segments) + '</speak>'
    return ssml

def process(text: str, rate: int, pitch: int, volume: int) -> str:
    if not text.strip():
        gr.Warning("لطفا متنی وارد کنید.")
        return ""

    ssml_text = build_ssml_from_dialog(text, default_voice="fa-IR-DilaraNeural")
    if not ssml_text:
        gr.Warning("متن نامعتبر است.")
        return ""

    # Convert sliders to strings only if needed
    rate_str = f"{rate}%" if rate != 0 else None
    pitch_str = f"{pitch}Hz" if pitch != 0 else None
    volume_str = f"{volume}%" if volume != 0 else None

    # Select first voice appearing in SSML
    first_voice = "fa-IR-DilaraNeural"
    if "fa-IR-FaridNeural" in ssml_text:
        first_voice = "fa-IR-FaridNeural"

    communicate = edge_tts.Communicate(
        text=ssml_text,
        voice=first_voice,
        rate=rate_str or "+0%",
        pitch=pitch_str or "+0Hz",
        volume=volume_str or "+0%",
    )

    temp_file = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
    temp_name = temp_file.name
    temp_file.close()

    # Important: ssml=True to parse voice tags
    communicate.save_sync(audio_fname=temp_name, ssml=True)

    return temp_name

def main():
    os.system("cls" if os.name == "nt" else "clear")

    text_area = gr.TextArea(
        lines=8,
        rtl=True,
        label="متن (از 'مرد:' و 'زن:' برای دیالوگ استفاده کنید)",
        placeholder="مثال:\nمرد: سلام\nزن: سلام، خوبی؟\nمرد: خوبم، ممنون.",
    )

    rate_slider = gr.Slider(label="سرعت", step=1, value=0, minimum=-100, maximum=100)
    pitch_slider = gr.Slider(label="فرکانس", step=1, value=0, minimum=-100, maximum=100)
    volume_slider = gr.Slider(label="حجم صدا", step=1, value=0, minimum=-100, maximum=100)

    result_audio = gr.Audio(label="نتیجه")

    interface = gr.Interface(
        fn=process,
        inputs=[text_area, rate_slider, pitch_slider, volume_slider],
        outputs=[result_audio],
        title="DT TTS (dialog support)",
        description="پشتیبانی از خطوط دیالوگ با برچسب 'مرد:' و 'زن:' — هر خط با صدای مربوط تولید می‌شود.",
        flagging_mode="never",
    )

    interface.launch()

if __name__ == "__main__":
    main()