File size: 11,381 Bytes
7934b29 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 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 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 |
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "RYGnI-EZp_nK"
},
"source": [
"# Getting Started: Sample Conversational AI application\n",
"This notebook shows how to use NVIDIA NeMo (https://github.com/NVIDIA/NeMo) to construct a toy demo which translate Mandarin audio file into English one.\n",
"\n",
"The demo demonstrates how to: \n",
"\n",
"* Instantiate pre-trained NeMo models from NVIDIA NGC.\n",
"* Transcribe audio with (Mandarin) speech recognition model.\n",
"* Translate text with machine translation model.\n",
"* Generate audio with text-to-speech models."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "V72HXYuQ_p9a"
},
"source": [
"## Installation\n",
"NeMo can be installed via simple pip command.\n",
"This will take about 4 minutes.\n",
"\n",
"(The installation method below should work inside your new Conda environment or in an NVIDIA docker container.)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "efDmTWf1_iYK"
},
"outputs": [],
"source": [
"BRANCH = 'r1.17.0'\n",
"!python -m pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[all]\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "EyJ5HiiPrPKA"
},
"source": [
"## Import all necessary packages"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "tdUqxeUEA8nw"
},
"outputs": [],
"source": [
"# Import NeMo and it's ASR, NLP and TTS collections\n",
"import nemo\n",
"# Import Speech Recognition collection\n",
"import nemo.collections.asr as nemo_asr\n",
"# Import Natural Language Processing colleciton\n",
"import nemo.collections.nlp as nemo_nlp\n",
"# Import Speech Synthesis collection\n",
"import nemo.collections.tts as nemo_tts\n",
"# We'll use this to listen to audio\n",
"import IPython"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "bt2EZyU3A1aq"
},
"source": [
"## Instantiate pre-trained NeMo models\n",
"\n",
"Every NeMo model has these methods:\n",
"\n",
"* ``list_available_models()`` - it will list all models currently available on NGC and their names.\n",
"\n",
"* ``from_pretrained(...)`` API downloads and initialized model directly from the NGC using model name.\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "YNNHs5Xjr8ox",
"scrolled": true
},
"outputs": [],
"source": [
"# Here is an example of all CTC-based models:\n",
"nemo_asr.models.EncDecCTCModel.list_available_models()\n",
"# More ASR Models are available - see: nemo_asr.models.ASRModel.list_available_models()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "1h9nhICjA5Dk",
"scrolled": true
},
"outputs": [],
"source": [
"# Speech Recognition model - Citrinet initially trained on Multilingual LibriSpeech English corpus, and fine-tuned on the open source Aishell-2\n",
"asr_model = nemo_asr.models.EncDecCTCModel.from_pretrained(model_name=\"stt_zh_citrinet_1024_gamma_0_25\").cuda()\n",
"\n",
"# Neural Machine Translation model\n",
"nmt_model = nemo_nlp.models.MTEncDecModel.from_pretrained(model_name='nmt_zh_en_transformer6x6').cuda()\n",
"\n",
"# Spectrogram generator which takes text as an input and produces spectrogram\n",
"spectrogram_generator = nemo_tts.models.FastPitchModel.from_pretrained(model_name=\"tts_en_fastpitch\").cuda()\n",
"\n",
"# Vocoder model which takes spectrogram and produces actual audio\n",
"vocoder = nemo_tts.models.HifiGanModel.from_pretrained(model_name=\"tts_en_hifigan\").cuda()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "KPota-JtsqSY"
},
"source": [
"## Get an audio sample in Mandarin"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "7cGCEKkcLr52"
},
"outputs": [],
"source": [
"# Download audio sample which we'll try\n",
"# This is a sample from MCV 6.1 Dev dataset - the model hasn't seen it before\n",
"# IMPORTANT: The audio must be mono with 16Khz sampling rate\n",
"audio_sample = 'common_voice_zh-CN_21347786.mp3'\n",
"!wget 'https://nemo-public.s3.us-east-2.amazonaws.com/zh-samples/common_voice_zh-CN_21347786.mp3'\n",
"# To listen it, click on the play button below\n",
"IPython.display.Audio(audio_sample)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "BaCdNJhhtBfM"
},
"source": [
"## Transcribe audio file\n",
"We will use speech recognition model to convert audio into text.\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "KTA7jM6sL6yC"
},
"outputs": [],
"source": [
"transcribed_text = asr_model.transcribe([audio_sample])\n",
"print(transcribed_text)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "BjYb2TMtttCc"
},
"source": [
"## Translate Chinese text into English\n",
"NeMo's NMT models have a handy ``.translate()`` method."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "kQTdE4b9Nm9O"
},
"outputs": [],
"source": [
"english_text = nmt_model.translate(transcribed_text)\n",
"print(english_text)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "9Rppc59Ut7uy"
},
"source": [
"## Generate English audio from text\n",
"Speech generation from text typically has two steps:\n",
"* Generate spectrogram from the text. In this example we will use FastPitch model for this.\n",
"* Generate actual audio from the spectrogram. In this example we will use HifiGan model for this.\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "wpMYfufgNt15"
},
"outputs": [],
"source": [
"# A helper function which combines FastPitch and HifiGan to go directly from \n",
"# text to audio\n",
"def text_to_audio(text):\n",
" parsed = spectrogram_generator.parse(text)\n",
" spectrogram = spectrogram_generator.generate_spectrogram(tokens=parsed)\n",
" audio = vocoder.convert_spectrogram_to_audio(spec=spectrogram)\n",
" return audio.to('cpu').detach().numpy()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Listen to generated audio in English\n",
"IPython.display.Audio(text_to_audio(english_text[0]), rate=22050)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "LiQ_GQpcBYUs"
},
"source": [
"## Next steps\n",
"A demo like this is great for prototyping and experimentation. However, for real production deployment, you would want to use a service like [NVIDIA Riva](https://developer.nvidia.com/riva).\n",
"\n",
"**NeMo is built for training.** You can fine-tune, or train from scratch on your data all models used in this example. We recommend you checkout the following, more in-depth, tutorials next:\n",
"\n",
"* [NeMo fundamentals](https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/00_NeMo_Primer.ipynb)\n",
"* [NeMo models](https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/01_NeMo_Models.ipynb)\n",
"* [Speech Recognition](https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/asr/ASR_with_NeMo.ipynb)\n",
"* [Punctuation and Capitalization](https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/nlp/Punctuation_and_Capitalization.ipynb)\n",
"* [Speech Synthesis](https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/tts/Inference_ModelSelect.ipynb)\n",
"\n",
"\n",
"You can find scripts for training and fine-tuning ASR, NLP and TTS models [here](https://github.com/NVIDIA/NeMo/tree/main/examples). "
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"name": "NeMo Getting Started",
"provenance": [],
"toc_visible": true
},
"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": "3.8.12"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
|