{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "33e4a305",
"metadata": {},
"outputs": [],
"source": [
"from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC, AutoModelForCTC, Wav2Vec2Processor\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"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "328d0662",
"metadata": {},
"outputs": [],
"source": [
"model = AutoModelForCTC.from_pretrained(\".\")\n",
"processor = Wav2Vec2Processor.from_pretrained(\".\")"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "0fea2518",
"metadata": {},
"outputs": [],
"source": [
"# model = AutoModelForCTC.from_pretrained(\"vitouphy/xls-r-300m-km\").to('cuda')\n",
"# processor = Wav2Vec2Processor.from_pretrained(\"vitouphy/xls-r-300m-km\")"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "9cfef23c",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using custom data configuration default-36119ec2a15afb82\n",
"Reusing dataset csv (/workspace/.cache/huggingface/datasets/csv/default-36119ec2a15afb82/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e)\n"
]
}
],
"source": [
"common_voice_test = (load_dataset('csv', data_files='km_kh_male/line_index_test.csv', split = 'train')\n",
" .remove_columns([\"Unnamed: 0\", \"drop\"])\n",
" .rename_column('text', 'sentence')\n",
" .cast_column(\"path\", Audio(sampling_rate=16_000)).rename_column('path', 'audio'))"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "29e6bb1a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'audio': {'path': '/workspace/xls-r-300m-km/km_kh_male/wavs/khm_3154_2555595821.wav',\n",
" 'array': array([ 0.00014737, 0.00016698, 0.00013704, ..., -0.00011244,\n",
" -0.0001059 , -0.00011476], dtype=float32),\n",
" 'sampling_rate': 16000},\n",
" 'sentence': 'ការ ធ្វើ អាជីវកម្ម រ៉ែ ដំបូង នៅ កម្ពុជា'}"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"common_voice_test[0]"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "0554b8d8",
"metadata": {},
"outputs": [],
"source": [
"def prepare_dataset(batch):\n",
" audio = batch[\"audio\"]\n",
" \n",
" # batched output is \"un-batched\"\n",
" batch[\"input_values\"] = processor(np.array(audio[\"array\"]), sampling_rate=audio[\"sampling_rate\"]).input_values[0]\n",
" batch[\"input_length\"] = len(batch[\"input_values\"])\n",
" \n",
" with processor.as_target_processor():\n",
" batch[\"labels\"] = processor(batch[\"sentence\"]).input_ids\n",
" return batch"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "d26a6659",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-36119ec2a15afb82/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-081703c0621182da.arrow\n"
]
}
],
"source": [
"common_voice_test = common_voice_test.map(prepare_dataset, remove_columns=common_voice_test.column_names)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "04a94f74",
"metadata": {},
"outputs": [],
"source": [
"i = 25"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "3993d2c4",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"It is strongly recommended to pass the ``sampling_rate`` argument to this function. Failing to do so can result in silent errors that might be hard to debug.\n"
]
}
],
"source": [
"input_dict = processor(common_voice_test[i][\"input_values\"], return_tensors=\"pt\", padding=True)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "7e3026dc",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'input_values': tensor([[ 2.8537e-04, 2.5043e-04, 2.7738e-04, ..., -4.8949e-05,\n",
" -1.1382e-04, 2.7166e-04]]), 'attention_mask': tensor([[1, 1, 1, ..., 1, 1, 1]], dtype=torch.int32)}"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"input_dict"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "adf215c0",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"It is strongly recommended to pass the ``sampling_rate`` argument to this function. Failing to do so can result in silent errors that might be hard to debug.\n"
]
}
],
"source": [
"input_dict = processor(common_voice_test[i][\"input_values\"], return_tensors=\"pt\", padding=True)\n",
"logits = model(input_dict.input_values.to(\"cuda\")).logits\n",
"pred_ids = torch.argmax(logits, dim=-1)[0]"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "e8310629",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([ 1, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72,\n",
" 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 10, 70, 70, 70, 10, 72,\n",
" 43, 72, 72, 72, 72, 72, 72, 0, 0, 72, 72, 18, 72, 54, 72, 72, 72, 72,\n",
" 72, 0, 72, 21, 72, 49, 72, 72, 72, 72, 72, 72, 23, 70, 70, 27, 72, 46,\n",
" 72, 72, 72, 1, 72, 0, 0, 30, 72, 72, 72, 72, 25, 70, 70, 72, 72, 11,\n",
" 55, 72, 72, 72, 72, 5, 72, 0, 20, 58, 72, 72, 72, 0, 0, 16, 72, 72,\n",
" 72, 20, 70, 70, 72, 72, 16, 70, 27, 72, 72, 72, 72, 72, 45, 0, 0, 30,\n",
" 30, 70, 70, 27, 72, 43, 72, 72, 72, 72, 72, 72, 21, 72, 53, 72, 72, 72,\n",
" 27, 72, 0, 1, 72, 72, 72, 72, 25, 70, 23, 23, 48, 72, 72, 72, 72, 72,\n",
" 72, 8, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72,\n",
" 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72,\n",
" 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72,\n",
" 72, 72, 72, 72, 72, 72, 72, 72, 43], device='cuda:0')"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"pred_ids"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "5dd986a0",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Prediction:\n",
"កញ្ញា ទេ បូព្រឹក សម្ដែង នៅ តន្ត្រី ស្រាបៀរ កម្ពុជា\n",
"\n",
"Reference:\n",
"កញ្ញា ទេព បូព្រឹក្ស សម្ដែង នៅ តន្ត្រី ស្រាបៀរ កម្ពុជា\n"
]
}
],
"source": [
"print(\"Prediction:\")\n",
"pred_ids = pred_ids[pred_ids != processor.tokenizer.pad_token_id]\n",
"print(processor.decode(pred_ids))\n",
"\n",
"print(\"\\nReference:\")\n",
"print(processor.decode(common_voice_test['labels'][i]))\n",
"# print(common_voice_test_transcription[0][\"sentence\"].lower())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8e39b112",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "562af933",
"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
}