{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "029da2d4",
      "metadata": {},
      "source": [
        "# OmniVoice — Lồng tiếng tiếng Việt trên Colab\n",
        "\n",
        "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/YOUR_USER/Omivoice/blob/main/hf/colab/Omivoice_VI_Colab.ipynb)\n",
        "\n",
        "Notebook chạy [OmniVoice](https://huggingface.co/k2-fsa/OmniVoice) với **6 giọng Việt** và pipeline **SRT lồng tiếng** (`native` — không cắt ngắt, tràn vào khoảng trống sau cue).\n",
        "\n",
        "**Yêu cầu:** Runtime **GPU** (T4+).\n",
        "\n",
        "**Trước khi chạy:** sửa `HF_REPO` thành dataset Hugging Face bạn đã upload."
      ]
    },
    {
      "cell_type": "markdown",
      "id": "26e11674",
      "metadata": {},
      "source": [
        "## 0. Cấu hình"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "899b610f",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Repo dataset trên HF — xem hf/README.md\n",
        "HF_REPO = \"STBack23/omnivoice-vi\"\n",
        "\n",
        "DEFAULT_VOICE = \"ngoc_huyen\"\n",
        "MERGE_MODE = \"native\"  # native | cascade | fit | strict\n",
        "NUM_STEP = 32"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "a96c5856",
      "metadata": {},
      "source": [
        "## 1. Kiểm tra GPU & cài đặt"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "52d2b7f3",
      "metadata": {},
      "outputs": [],
      "source": [
        "import torch\n",
        "assert torch.cuda.is_available(), \"Bật GPU: Runtime → Change runtime type → T4 GPU\"\n",
        "print(f\"GPU: {torch.cuda.get_device_name(0)}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ef948472",
      "metadata": {},
      "outputs": [],
      "source": [
        "!pip install -q omnivoice audiotsm huggingface_hub soundfile librosa"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "12ea36ff",
      "metadata": {},
      "source": [
        "## 2. Tải giọng & công cụ từ Hugging Face"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ef3f54c5",
      "metadata": {},
      "outputs": [],
      "source": [
        "from pathlib import Path\n",
        "from huggingface_hub import snapshot_download\n",
        "\n",
        "ROOT = Path(snapshot_download(HF_REPO, repo_type=\"dataset\", local_dir=\"/content/omnivoice-vi\"))\n",
        "print(f\"Đã tải: {ROOT}\")\n",
        "print(\"Giọng:\", [p.parent.name for p in sorted((ROOT / \"voices\").glob(\"*/profile.json\"))])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "86a2c0ac",
      "metadata": {},
      "outputs": [],
      "source": [
        "import importlib.util\n",
        "import sys\n",
        "\n",
        "speak_path = ROOT / \"tools\" / \"speak.py\"\n",
        "spec = importlib.util.spec_from_file_location(\"speak\", speak_path)\n",
        "speak = importlib.util.module_from_spec(spec)\n",
        "sys.modules[\"speak\"] = speak\n",
        "spec.loader.exec_module(speak)\n",
        "print(\"Đã load speak.py\")"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "2f09d0f9",
      "metadata": {},
      "source": [
        "## 3. Nạp model (~2–3 phút lần đầu)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ac82e5d7",
      "metadata": {},
      "outputs": [],
      "source": [
        "PROFILE = speak.load_profile(ROOT / \"voices\" / DEFAULT_VOICE / \"profile.json\")\n",
        "MODEL = speak.load_model(PROFILE)\n",
        "PROMPT = speak.ensure_voice_prompt(MODEL, PROFILE)\n",
        "print(f\"Giọng: {PROFILE['name']} | SR: {MODEL.sampling_rate}\")"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "52684ad7",
      "metadata": {},
      "source": [
        "## 4. Đọc thử một câu"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "657ecf07",
      "metadata": {},
      "outputs": [],
      "source": [
        "import soundfile as sf\n",
        "from IPython.display import Audio, display\n",
        "\n",
        "TEXT = \"Xin chào, đây là thử nghiệm lồng tiếng tiếng Việt bằng OmniVoice trên Colab.\"\n",
        "audio = speak.synthesize(MODEL, PROMPT, TEXT, PROFILE[\"language\"], num_step=NUM_STEP)\n",
        "sf.write(\"/content/demo.wav\", audio, MODEL.sampling_rate)\n",
        "display(Audio(\"/content/demo.wav\"))"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "e037e417",
      "metadata": {},
      "source": [
        "## 5. Đổi giọng"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "67b577c2",
      "metadata": {},
      "outputs": [],
      "source": [
        "def load_voice(slug: str):\n",
        "    global PROFILE, PROMPT\n",
        "    PROFILE = speak.load_profile(ROOT / \"voices\" / slug / \"profile.json\")\n",
        "    PROMPT = speak.ensure_voice_prompt(MODEL, PROFILE)\n",
        "    print(f\"Đã chọn: {PROFILE['name']} ({slug})\")\n",
        "\n",
        "load_voice(\"ban_mai\")"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "ed3c3013",
      "metadata": {},
      "source": [
        "## 6. Lồng tiếng file SRT\n",
        "\n",
        "Upload `.srt`, ghép WAV với mode `native`."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "33f7fbc1",
      "metadata": {},
      "outputs": [],
      "source": [
        "from google.colab import files\n",
        "\n",
        "print(\"Upload file .srt:\")\n",
        "uploaded = files.upload()\n",
        "SRT_PATH = Path(\"/content\") / list(uploaded.keys())[0]\n",
        "print(f\"File: {SRT_PATH}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "dcc05960",
      "metadata": {},
      "outputs": [],
      "source": [
        "profile_path = ROOT / \"voices\" / PROFILE[\"slug\"] / \"profile.json\"\n",
        "\n",
        "result = speak.run_srt_pipeline(\n",
        "    profile_path,\n",
        "    SRT_PATH,\n",
        "    merge=True,\n",
        "    merge_mode=MERGE_MODE,\n",
        "    num_step=NUM_STEP,\n",
        "    model=MODEL,\n",
        "    prompt=PROMPT,\n",
        ")\n",
        "\n",
        "print(result[\"log\"])\n",
        "merged = result.get(\"merged_wav\")\n",
        "if merged:\n",
        "    display(Audio(merged))\n",
        "    files.download(merged)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "a2e288e3",
      "metadata": {},
      "source": [
        "## 7. Clone giọng mới (tùy chọn)\n",
        "\n",
        "Upload audio mẫu 3–10 giây. `REF_TEXT = None` để Whisper tự nhận."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "09ec9960",
      "metadata": {},
      "outputs": [],
      "source": [
        "print(\"Upload ref audio (wav/mp3):\")\n",
        "ref_up = files.upload()\n",
        "REF_AUDIO = Path(\"/content\") / list(ref_up.keys())[0]\n",
        "REF_TEXT = None\n",
        "\n",
        "clone_prompt = MODEL.create_voice_clone_prompt(\n",
        "    ref_audio=str(REF_AUDIO),\n",
        "    ref_text=REF_TEXT,\n",
        "    preprocess_prompt=True,\n",
        ")\n",
        "clone_audio = speak.synthesize(\n",
        "    MODEL, clone_prompt,\n",
        "    \"Đây là giọng clone từ audio mẫu bạn vừa upload.\",\n",
        "    \"Vietnamese\", num_step=NUM_STEP,\n",
        ")\n",
        "sf.write(\"/content/clone.wav\", clone_audio, MODEL.sampling_rate)\n",
        "display(Audio(\"/content/clone.wav\"))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## 8. Giải phóng VRAM"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "del MODEL, PROMPT\n",
        "import gc\n",
        "gc.collect()\n",
        "torch.cuda.empty_cache()\n",
        "print(\"Đã giải phóng VRAM\")"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "gpuType": "T4",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.10.0"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}