{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5393aa33",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC, AutoModelForCTC, Wav2Vec2Processor, AutoProcessor, Wav2Vec2ProcessorWithLM\n",
    "from datasets import load_dataset, load_metric, Audio\n",
    "from pyctcdecode import build_ctcdecoder\n",
    "from pydub import AudioSegment\n",
    "from pydub.playback import play\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "import kenlm\n",
    "import pandas as pd\n",
    "import random\n",
    "import soundfile as sf\n",
    "from tqdm.auto import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2d34d3b8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# KENLM_MODEL_LOC = '/workspace/xls-r-300m-km/data/km_text_word_unigram.arpa'\n",
    "KENLM_MODEL_LOC = '/workspace/xls-r-300m-km/data/km_wiki_ngram.arpa'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f0354cb2",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading the LM will be faster if you build a binary file.\n",
      "Reading /workspace/xls-r-300m-km/vitouphy/xls-r-300m-km/language_model/km_text.arpa\n",
      "----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100\n",
      "Only 81 unigrams passed as vocabulary. Is this small or artificial data?\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "processor = AutoProcessor.from_pretrained(\"vitouphy/xls-r-300m-km\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "109f28e9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'|': 0, 'ក': 1, 'ខ': 2, 'គ': 3, 'ឃ': 4, 'ង': 5, 'ច': 6, 'ឆ': 7, 'ជ': 8, 'ឈ': 9, 'ញ': 10, 'ដ': 11, 'ឋ': 12, 'ឌ': 13, 'ឍ': 14, 'ណ': 15, 'ត': 16, 'ថ': 17, 'ទ': 18, 'ធ': 19, 'ន': 20, 'ប': 21, 'ផ': 22, 'ព': 23, 'ភ': 24, 'ម': 25, 'យ': 26, 'រ': 27, 'ល': 28, 'វ': 29, 'ស': 30, 'ហ': 31, 'ឡ': 32, 'អ': 33, 'ឥ': 34, 'ឧ': 35, 'ឪ': 36, 'ឫ': 37, 'ឬ': 38, 'ឭ': 39, 'ឮ': 40, 'ឯ': 41, 'ឱ': 42, 'ា': 43, 'ិ': 44, 'ី': 45, 'ឹ': 46, 'ឺ': 47, 'ុ': 48, 'ូ': 49, 'ួ': 50, 'ើ': 51, 'ឿ': 52, 'ៀ': 53, 'េ': 54, 'ែ': 55, 'ៃ': 56, 'ោ': 57, 'ៅ': 58, 'ំ': 59, 'ះ': 60, 'ៈ': 61, '៉': 62, '៊': 63, '់': 64, '៌': 65, '៍': 66, '៎': 67, '៏': 68, '័': 69, '្': 70, '[unk]': 71, '[pad]': 72, '<s>': 73, '</s>': 74}\n"
     ]
    }
   ],
   "source": [
    "vocab_dict = processor.tokenizer.get_vocab()\n",
    "sorted_vocab_dict = {k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])}\n",
    "print(sorted_vocab_dict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "300cec39",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading the LM will be faster if you build a binary file.\n",
      "Reading /workspace/xls-r-300m-km/data/km_wiki_ngram.arpa\n",
      "----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100\n",
      "Found entries of length > 1 in alphabet. This is unusual unless style is BPE, but the alphabet was not recognized as BPE type. Is this correct?\n",
      "****************************************************************************************************\n"
     ]
    }
   ],
   "source": [
    "decoder = build_ctcdecoder(\n",
    "    labels=list(sorted_vocab_dict.keys()),\n",
    "    kenlm_model_path=KENLM_MODEL_LOC,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "27dd8427",
   "metadata": {},
   "outputs": [],
   "source": [
    "processor_with_lm = Wav2Vec2ProcessorWithLM(\n",
    "    feature_extractor=processor.feature_extractor,\n",
    "    tokenizer=processor.tokenizer,\n",
    "    decoder=decoder\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "94eb248e",
   "metadata": {},
   "outputs": [],
   "source": [
    "processor_with_lm.save_pretrained(\".\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8f9b3dcc",
   "metadata": {},
   "source": [
    "## Save Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8b584690",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bc5bf68946064e97b869d44b02e7af19",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/1.18G [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model = AutoModelForCTC.from_pretrained(\"vitouphy/xls-r-300m-km\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "3712c030",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_pretrained('.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b5d8de20",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}