{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "Training_Example.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "2zx09MSGFHjT"
      },
      "source": [
        "!pip install deep-phonemizer"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oZ1L1BlhOUMR"
      },
      "source": [
        "# Dowload and prepare a dataset\n",
        "!wget https://raw.githubusercontent.com/CUNY-CL/wikipron/master/data/scrape/tsv/eng_latn_us_broad.tsv\n",
        "\n",
        "with open('eng_latn_us_broad.tsv', 'r', encoding='utf-8') as f:\n",
        "  lines = f.readlines()\n",
        "\n",
        "# Prepare data as tuples (lang, word, phoneme)\n",
        "lines = [l.replace(' ', '').replace('\\n', '') for l in lines]\n",
        "splits = [l.split('\\t') for l in lines]\n",
        "train_data = [('en_us', s[0], s[1]) for s in splits if len(s)==2]\n",
        "\n",
        "for d in train_data[:10000:1000]:\n",
        "  print(d)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MiV-OUi0FQ-O"
      },
      "source": [
        "# Read standard config and adjust some params for speedup\n",
        "from dp.utils.io import read_config, save_config\n",
        "import dp\n",
        "import os\n",
        "\n",
        "config_file = os.path.dirname(dp.__file__) + '/configs/forward_config.yaml'\n",
        "config = read_config(config_file)\n",
        "config['training']['epochs'] = 10\n",
        "config['training']['warmup_steps'] = 100\n",
        "config['training']['generate_steps'] = 500\n",
        "config['training']['validate_steps'] = 500\n",
        "save_config(config, 'config.yaml')\n",
        "\n",
        "for k, v in config.items():\n",
        "  print(f'{k} {v}')\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "m_SNVv_HN4fR"
      },
      "source": [
        "%load_ext tensorboard\n",
        "%tensorboard --logdir /content/checkpoints"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K09wG-ymXECZ"
      },
      "source": [
        "from dp.preprocess import preprocess\n",
        "from dp.train import train\n",
        "\n",
        "preprocess(config_file='config.yaml', train_data=train_data)\n",
        "train(config_file='config.yaml')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2KcIwL6QdvEJ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ed499929-1103-4a22-dce8-4a54d85f9b9e"
      },
      "source": [
        "# Load phonemizer (including the training data dictionary)\n",
        "from dp.phonemizer import Phonemizer\n",
        "\n",
        "phonemizer = Phonemizer.from_checkpoint('/content/checkpoints/best_model.pt')\n",
        "result = phonemizer('Phonemizing an English text is imposimpable!', lang='en_us')\n",
        "\n",
        "print(result)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2021-05-12 10:15:19,916.916 DEBUG phonemizer:  Initializing phonemizer with model step 18000\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "fənəmaɪzɪŋ ən ɪŋɡlɪʃ tɛkst ɪz ɪmpɑsɪmpəbəl!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xt85fzFneDno",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bc1cb892-9cd2-4540-b930-e4443ddf69c2"
      },
      "source": [
        "# Phonemize a list of texts and pull out model predictions with confidence scores\n",
        "result = phonemizer.phonemise_list(['Phonemizing an US-English text is imposimpable!'], lang='en_us')\n",
        "\n",
        "for word, pred in result.predictions.items():\n",
        "  print(f'{word} {pred.phonemes} {pred.confidence}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "imposimpable ɪmpɑsɪmpəbəl 0.2185952042855603\n",
            "Phonemizing fənəmaɪzɪŋ 0.22222847233670942\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}