{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 🗣️ Lahgtna — Egyptian-Arabic TTS on Colab\n",
    "\n",
    "Run the **v3** model on a free Colab GPU.\n",
    "**First: Runtime → Change runtime type → T4 GPU.**"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "!pip install -q omnivoice catt-tashkeel num2words soundfile"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Log in to Hugging Face (private repo — use a READ token)"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "from huggingface_hub import login\n",
    "login()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load model + reference voice"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "import torch\n",
    "from omnivoice.models.omnivoice import OmniVoice\n",
    "from huggingface_hub import hf_hub_download\n",
    "\n",
    "REPO = \"ehabnegm/lahgtna-omnivoice-egyptian-v3\"\n",
    "model = OmniVoice.from_pretrained(REPO, device_map=\"cuda\", dtype=torch.float16)\n",
    "ref_audio = hf_hub_download(REPO, \"reference.wav\")\n",
    "ref_text  = \"كان العمل التطوعي واللي لما تفتح الباب بس ليه الناس\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Text front-end + long-form helper (chunk → no drift)"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "import re, numpy as np, soundfile as sf\n",
    "from num2words import num2words\n",
    "_DIGITS = str.maketrans(\"٠١٢٣٤٥٦٧٨٩\", \"0123456789\")\n",
    "HARAKAT = re.compile(r\"[ً-ْٰـ]\")\n",
    "def prep(t):\n",
    "    t = t.translate(_DIGITS)\n",
    "    t = re.sub(r\"\\d+\", lambda m: \" \" + num2words(int(m.group()), lang=\"ar\") + \" \", t)\n",
    "    t = re.sub(r\"[A-Za-z]+\", \" \", t)\n",
    "    t = HARAKAT.sub(\"\", t).replace(\"ى\", \"ي\")\n",
    "    return re.sub(r\"\\s+\", \" \", t).strip()\n",
    "def chunks(t):\n",
    "    out, cur = [], \"\"\n",
    "    for p in re.split(r\"([،؛\\.؟!\\n]+)\", t):\n",
    "        cur += p\n",
    "        if re.search(r\"[،؛\\.؟!\\n]\", p):\n",
    "            if cur.strip(): out.append(cur.strip())\n",
    "            cur = \"\"\n",
    "    if cur.strip(): out.append(cur.strip())\n",
    "    return out or [t]\n",
    "def synth(text, num_step=16):\n",
    "    parts = [model.generate(text=prep(c), language=\"arz\", ref_audio=ref_audio, ref_text=ref_text, num_step=num_step)[0] for c in chunks(text) if prep(c)]\n",
    "    return np.concatenate(parts)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate & play"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "from IPython.display import Audio\n",
    "text = \"أهلاً بحضرتك، معاك المساعد الذكي. اتفضل قوللي محتاج إيه؟\"\n",
    "sf.write(\"out.wav\", synth(text), 24000)\n",
    "Audio(\"out.wav\")"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "provenance": []
  },
  "kernelspec": {
   "name": "python3",
   "display_name": "Python 3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}