Add colab training notebook

notebooks/Piper_Training.ipynb

@@ -0,0 +1,350 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "provenance": [],
+      "gpuType": "T4",
+      "authorship_tag": "ABX9TyMLK8+OPF/URaEggfgBAHja",
+      "include_colab_link": true
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    },
+    "accelerator": "GPU"
+  },
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "view-in-github",
+        "colab_type": "text"
+      },
+      "source": [
+        "<a href=\"https://colab.research.google.com/github/natlamir/piper1-gpl/blob/add-colab-training-notebook/notebooks/Piper_Training.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "0) Colab runtime + GPU check"
+      ],
+      "metadata": {
+        "id": "ulTkpdiGQjI4"
+      }
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "Y37QtlLmPuqT"
+      },
+      "outputs": [],
+      "source": [
+        "# Check GPU\n",
+        "import torch, platform, sys\n",
+        "print(\"Python:\", sys.version.split()[0])\n",
+        "print(\"PyTorch:\", torch.__version__)\n",
+        "print(\"CUDA available:\", torch.cuda.is_available())\n",
+        "!nvidia-smi"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "1) System packages (incl. eSpeak dev)"
+      ],
+      "metadata": {
+        "id": "tRWDRJUpQrfw"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!sudo apt-get update -y\n",
+        "!sudo apt-get install -y build-essential cmake ninja-build espeak-ng espeak-ng-data libespeak-ng-dev pkg-config ffmpeg\n",
+        "!pkg-config --modversion espeak-ng"
+      ],
+      "metadata": {
+        "id": "aN5lcQRZQs3D"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "2) Clone repo fresh"
+      ],
+      "metadata": {
+        "id": "J_MRPygaQ1cQ"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "%cd /content\n",
+        "!rm -rf piper1-gpl\n",
+        "!git clone https://github.com/OHF-voice/piper1-gpl.git\n",
+        "%cd piper1-gpl\n",
+        "!pwd"
+      ],
+      "metadata": {
+        "id": "w-0x1IWPQ7UB"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "3) Python deps (editable install, no venv in Colab)"
+      ],
+      "metadata": {
+        "id": "k_qVUXtoRArT"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!python3 -m pip install --upgrade pip setuptools wheel\n",
+        "!python3 -m pip install -e \".[train]\""
+      ],
+      "metadata": {
+        "id": "L9ARM5C9RDZ_"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "4) Build the Cython extension used for alignment"
+      ],
+      "metadata": {
+        "id": "OXkhMJRkRUGk"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "%cd /content/piper1-gpl\n",
+        "!chmod +x ./build_monotonic_align.sh\n",
+        "!./build_monotonic_align.sh"
+      ],
+      "metadata": {
+        "id": "EMQF3quvRVe7"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "5) Dev build (repo mode)"
+      ],
+      "metadata": {
+        "id": "f8P9iQhjRa81"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!python3 -m pip install --upgrade pip setuptools wheel scikit-build cmake ninja"
+      ],
+      "metadata": {
+        "id": "ZIcRcqt6p3kM"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "%cd /content/piper1-gpl\n",
+        "!python3 setup.py build_ext --inplace -v"
+      ],
+      "metadata": {
+        "id": "bTNUkZKmRcJb"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "6) (Optional) Mount Google Drive for datasets and outputs"
+      ],
+      "metadata": {
+        "id": "xsmEOdPpRjPo"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from google.colab import drive\n",
+        "drive.mount('/content/drive')"
+      ],
+      "metadata": {
+        "id": "S82oT9EaRlR0"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "7) Set paths and training hyperparams"
+      ],
+      "metadata": {
+        "id": "bRewWxitSmSr"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from pathlib import Path\n",
+        "\n",
+        "# ==== CHANGE THESE ====\n",
+        "VOICE_NAME      = \"my_colab_voice\"\n",
+        "ESPEAK_VOICE    = \"en-us\"  # run `!espeak-ng --voices` to see options\n",
+        "SAMPLE_RATE_HZ  = 22050\n",
+        "BATCH_SIZE      = 32       # drop to 8 or 4 if you OOM\n",
+        "\n",
+        "DATA_ROOT       = Path(\"/content/drive/MyDrive/tts_data/myvoice\")\n",
+        "AUDIO_DIR       = DATA_ROOT / \"wavs\"\n",
+        "CSV_PATH        = DATA_ROOT / \"metadata.csv\"\n",
+        "\n",
+        "CACHE_DIR       = Path(\"/content/piper_cache\")\n",
+        "CONFIG_PATH     = DATA_ROOT / f\"{VOICE_NAME}.json\"\n",
+        "\n",
+        "# Optional: start from an existing checkpoint to speed up & stabilize training\n",
+        "# Get a .ckpt from https://huggingface.co/datasets/rhasspy/piper-checkpoints (medium quality recommended)\n",
+        "CKPT_PATH       = \"\"  # e.g., \"/content/drive/MyDrive/piper_ckpts/en_US-lessac-medium.ckpt\"\n",
+        "\n",
+        "# Make sure dirs exist\n",
+        "CACHE_DIR.mkdir(parents=True, exist_ok=True)\n",
+        "print(\"CSV exists:\", CSV_PATH.exists())\n",
+        "print(\"Audio dir exists:\", AUDIO_DIR.exists())\n",
+        "print(\"Cache dir:\", CACHE_DIR)\n",
+        "print(\"Config will be written to:\", CONFIG_PATH)"
+      ],
+      "metadata": {
+        "id": "nyXNxSWFSq2u"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "8) Quick sanity checks"
+      ],
+      "metadata": {
+        "id": "YZqVTzcDSwRV"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!espeak-ng --voices | head -n 20"
+      ],
+      "metadata": {
+        "id": "PPTTUbbxSydH"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import pandas as pd, io, os, textwrap\n",
+        "\n",
+        "csv_path = str(CSV_PATH)\n",
+        "if os.path.exists(csv_path):\n",
+        "    # Read as pipe-delimited, two columns\n",
+        "    try:\n",
+        "        df = pd.read_csv(csv_path, sep=\"|\", header=None, names=[\"audio\",\"text\"])\n",
+        "        print(df.head())\n",
+        "        # Check a few audio files exist\n",
+        "        missing = [a for a in df[\"audio\"].head(5) if not (AUDIO_DIR/str(a)).exists()]\n",
+        "        print(\"Missing among first 5:\", missing)\n",
+        "    except Exception as e:\n",
+        "        print(\"CSV read error:\", e)\n",
+        "else:\n",
+        "    print(\"CSV not found at:\", csv_path)"
+      ],
+      "metadata": {
+        "id": "JJImgsKzS1pk"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "9) Kick off training"
+      ],
+      "metadata": {
+        "id": "jiggGLHzTxGI"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!python3 -m piper.train fit \\\n",
+        "  --data.voice_name \"my_colab_voice\" \\\n",
+        "  --data.csv_path \"/content/drive/MyDrive/tts_data/myvoice/metadata.csv\" \\\n",
+        "  --data.audio_dir \"/content/drive/MyDrive/tts_data/myvoice/wavs\" \\\n",
+        "  --model.sample_rate 22050 \\\n",
+        "  --data.espeak_voice \"en-us\" \\\n",
+        "  --data.cache_dir \"/content/piper_cache\" \\\n",
+        "  --data.config_path \"/content/drive/MyDrive/tts_data/myvoice/my_colab_voice.json\" \\\n",
+        "  --data.batch_size 8 \\\n",
+        "  --ckpt_path \"https://huggingface.co/datasets/rhasspy/piper-checkpoints/resolve/main/en/en_US/lessac/medium/epoch%3D2164-step%3D1355540.ckpt\""
+      ],
+      "metadata": {
+        "id": "PZigA3tIUSeh"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "10. Export to onnx"
+      ],
+      "metadata": {
+        "id": "bSUfANcq7jUX"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!python3 -m piper.train.export_onnx \\\n",
+        "  --checkpoint \"/content/piper1-gpl/lightning_logs/version_2/checkpoints/epoch=2174-step=2680.ckpt\" \\\n",
+        "  --output-file \"/content/model.onnx\""
+      ],
+      "metadata": {
+        "id": "oX63f32Q7m98"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!cp /content/drive/MyDrive/tts_data/myvoice/my_colab_voice.json /content/model.onnx.json"
+      ],
+      "metadata": {
+        "id": "pOqvuxqM8UrN"
+      },
+      "execution_count": null,
+      "outputs": []
+    }
+  ]
+}