app.py

"""
app.py – Gradio TTS Space (Arabic only)
يدعم: Piper · Supertonic-3 · Chatterbox Multilingual
"""
import sys
print("=== app.py starting ===", flush=True)

# ── import check مبكر عشان نشوف أي مشكلة قبل Gradio ──────────────────────
_import_errors = []

try:
    import onnxruntime
    print(f"[ok] onnxruntime {onnxruntime.__version__}", flush=True)
except Exception as e:
    _import_errors.append(f"onnxruntime: {e}")

try:
    from piper import PiperVoice
    print("[ok] piper", flush=True)
except Exception as e:
    _import_errors.append(f"piper: {e}")

try:
    from supertonic import TTS as _ST
    print("[ok] supertonic", flush=True)
except Exception as e:
    _import_errors.append(f"supertonic: {e}")

try:
    from engines import generate_audio, MODELS_CONFIG
    print("[ok] engines", flush=True)
except Exception as e:
    _import_errors.append(f"engines: {e}")
    print(f"[FATAL] engines import failed: {e}", flush=True)
    sys.exit(1)

if _import_errors:
    print("[WARNING] بعض الـ imports فشلت:", flush=True)
    for err in _import_errors:
        print(f"  - {err}", flush=True)

# ──────────────────────────────────────────────────────────────────────────
import os
import wave
import tempfile
import gradio as gr

MODEL_DESCRIPTIONS = {
    "Piper – ar (Arabic)":          "🎙️ Piper VITS · ONNX CPU · عربي فصحى (أردني)",
    "Supertonic-3 – M1 (Arabic)":  "⚡ Supertonic-3 · صوت ذكر · عربي",
    "Supertonic-3 – F1 (Arabic)":  "⚡ Supertonic-3 · صوت أنثى · عربي",
    "Chatterbox Multilingual – ar": "🔊 Chatterbox ML · ONNX · عربي",
}

MODEL_NAMES = list(MODELS_CONFIG.keys())


def update_desc(model_name):
    return MODEL_DESCRIPTIONS.get(model_name, "")


def run_tts(model_name, text):
    if not model_name:
        return None, "❌ اختر موديل أولاً"
    if not text or not text.strip():
        return None, "❌ اكتب نصاً أولاً"

    try:
        tmp      = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
        out_path = tmp.name
        tmp.close()

        print(f"[run_tts] model={model_name!r} text={text[:60]!r}", flush=True)

        generate_audio(model_name, text, out_path)

        exists = os.path.exists(out_path)
        size   = os.path.getsize(out_path) if exists else -1
        print(f"[run_tts] exists={exists} size={size}", flush=True)

        if not exists or size <= 44:
            return None, f"❌ الملف فارغ أو غير موجود (size={size})"

        # فحص WAV (Piper بيعمل WAV، Chatterbox بيعمل WAV عبر soundfile)
        try:
            with wave.open(out_path, "r") as wf:
                frames = wf.getnframes()
                rate   = wf.getframerate()
                print(f"[run_tts] WAV frames={frames} rate={rate} dur={frames/rate:.2f}s", flush=True)
                if frames == 0:
                    return None, "❌ WAV header موجود بس مفيش audio — راجع terminal"
        except Exception:
            # soundfile بيكتب WAV بشكل مختلف أحياناً، متقلقش لو wave.open فشل
            pass

        return out_path, f"✅ تم التوليد ({size // 1024} KB)"

    except Exception as e:
        import traceback
        print(f"[run_tts] ERROR:\n{traceback.format_exc()}", flush=True)
        return None, f"❌ خطأ: {e}"


# ── Gradio UI ──────────────────────────────────────────────────────────────
with gr.Blocks(
    title="🎙️ TTS عربي – Piper · Supertonic-3 · Chatterbox",
    theme=gr.themes.Soft(primary_hue="indigo"),
) as demo:

    gr.Markdown("""
    # 🎙️ مقارنة نماذج TTS العربية
    **Piper · Supertonic-3 · Chatterbox Multilingual**

    كل النماذج تعمل على CPU عبر ONNX Runtime.
    أول تشغيل لكل نموذج يحمّله من HuggingFace تلقائياً.
    """)

    with gr.Row():
        with gr.Column(scale=1):
            model_dropdown = gr.Dropdown(
                choices=MODEL_NAMES,
                value=MODEL_NAMES[0],
                label="🤖 اختر النموذج",
            )
            model_desc = gr.Markdown(
                value=MODEL_DESCRIPTIONS.get(MODEL_NAMES[0], ""),
            )
            text_input = gr.Textbox(
                label="📝 النص العربي",
                placeholder="اكتب النص هنا …",
                lines=5,
                rtl=True,
            )
            generate_btn = gr.Button("🔊 توليد الصوت", variant="primary")

        with gr.Column(scale=1):
            audio_out = gr.Audio(
                label="🎵 الصوت المولّد",
                type="filepath",
                show_download_button=True,
                interactive=False,
            )
            status_out = gr.Markdown(value="")

    model_dropdown.change(
        fn=update_desc,
        inputs=[model_dropdown],
        outputs=[model_desc],
    )
    generate_btn.click(
        fn=run_tts,
        inputs=[model_dropdown, text_input],
        outputs=[audio_out, status_out],
    )

    gr.Markdown("""
    ---
    ### 📦 ملاحظات
    - **Piper**: يحتاج `espeak-ng` مثبّتاً على النظام
    - **Supertonic-3**: ~400MB عند أول تحميل
    - **Chatterbox ML**: ~2GB عند أول تحميل
    """)


if __name__ == "__main__":
    demo.launch()