File size: 16,666 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
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
{
    "cells": [
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "# TTS Inference Model Selection\n",
                "\n",
                "This notebook can be used to generate audio samples using either NeMo's pretrained models or after training NeMo TTS models. This notebook supports all TTS models and is intended to showcase different models and how their results differ."
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "# License\n",
                "\n",
                "> Copyright 2020 NVIDIA. All Rights Reserved.\n",
                "> \n",
                "> Licensed under the Apache License, Version 2.0 (the \"License\");\n",
                "> you may not use this file except in compliance with the License.\n",
                "> You may obtain a copy of the License at\n",
                "> \n",
                ">     http://www.apache.org/licenses/LICENSE-2.0\n",
                "> \n",
                "> Unless required by applicable law or agreed to in writing, software\n",
                "> distributed under the License is distributed on an \"AS IS\" BASIS,\n",
                "> WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
                "> See the License for the specific language governing permissions and\n",
                "> limitations under the License."
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {
                "tags": []
            },
            "outputs": [],
            "source": [
                "\"\"\"\n",
                "You can either run this notebook locally (if you have all the dependencies and a GPU) or on Google Colab.\n",
                "Instructions for setting up Colab are as follows:\n",
                "1. Open a new Python 3 notebook.\n",
                "2. Import this notebook from GitHub (File -> Upload Notebook -> \"GITHUB\" tab -> copy/paste GitHub URL)\n",
                "3. Connect to an instance with a GPU (Runtime -> Change runtime type -> select \"GPU\" for hardware accelerator)\n",
                "4. Run this cell to set up dependencies.\n",
                "\"\"\"\n",
                "BRANCH = 'r1.17.0'\n",
                "# # If you're using Google Colab and not running locally, uncomment and run this cell.\n",
                "# !apt-get install sox libsndfile1 ffmpeg\n",
                "# !pip install wget text-unidecode\n",
                "# !python -m pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[all]\n"
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "## Models\n",
                "\n",
                "First we pick the models that we want to use. Currently supported models are:\n",
                "\n",
                "Spectrogram Generators:\n",
                "- [Tacotron 2](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_tacotron2)\n",
                "- [FastPitch](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_fastpitch)\n",
                "- [Mixer-TTS](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_lj_mixertts)\n",
                "- [Mixer-TTS-X](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_lj_mixerttsx)\n",
                "\n",
                "Audio Generators\n",
                "- [WaveGlow](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_waveglow_88m)\n",
                "- [HiFiGAN](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_hifigan)\n",
                "- [UnivNet](https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_lj_univnet)\n",
                "- Griffin-Lim"
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {
                "tags": []
            },
            "outputs": [],
            "source": [
                "from ipywidgets import Select, HBox, Label\n",
                "from IPython.display import display\n",
                "\n",
                "supported_spec_gen = [\"tacotron2\", \"fastpitch\", \"mixertts\", \"mixerttsx\", None]\n",
                "supported_audio_gen = [\"waveglow\", \"hifigan\", \"univnet\", \"griffin-lim\", None]\n",
                "\n",
                "print(\"Select the model(s) that you want to use. Please choose 1 spectrogram generator and 1 vocoder.\")\n",
                "spectrogram_generator_selector = Select(options=supported_spec_gen, value=None)\n",
                "audio_generator_selector = Select(options=supported_audio_gen, value=None)\n",
                "display(HBox([spectrogram_generator_selector, Label(\"+\"), audio_generator_selector]))"
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {},
            "outputs": [],
            "source": [
                "spectrogram_generator = spectrogram_generator_selector.value\n",
                "audio_generator = audio_generator_selector.value\n",
                "\n",
                "if spectrogram_generator is None and audio_generator is None:\n",
                "    raise ValueError(\"No models were chosen. Please return to the previous step and choose either 1 end-to-end model or 1 spectrogram generator and 1 vocoder.\")\n",
                "\n",
                "if (spectrogram_generator and audio_generator is None) or (audio_generator and spectrogram_generator is None):\n",
                "    raise ValueError(\"In order to continue with the two step pipeline, both the spectrogram generator and the audio generator must be chosen, but one was `None`\")"
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "## Load model checkpoints\n",
                "\n",
                "Next we load the pretrained model provided by NeMo. All NeMo models have two functions to help with this\n",
                "\n",
                "- list_available_models(): This function will return a list of all pretrained checkpoints for that model\n",
                "- from_pretrained(): This function will download the pretrained checkpoint, load it, and return an instance of the model\n",
                "\n",
                "Below we will use `from_pretrained` to load the chosen models from above."
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {
                "scrolled": false,
                "tags": []
            },
            "outputs": [],
            "source": [
                "from omegaconf import OmegaConf, open_dict\n",
                "import torch\n",
                "from nemo.collections.tts.models.base import SpectrogramGenerator, Vocoder\n",
                "\n",
                "\n",
                "def load_spectrogram_model():\n",
                "    override_conf = None\n",
                "    \n",
                "    from_pretrained_call = SpectrogramGenerator.from_pretrained\n",
                "    \n",
                "    if spectrogram_generator == \"tacotron2\":\n",
                "        from nemo.collections.tts.models import Tacotron2Model\n",
                "        pretrained_model = \"tts_en_tacotron2\"\n",
                "    elif spectrogram_generator == \"fastpitch\":\n",
                "        from nemo.collections.tts.models import FastPitchModel\n",
                "        pretrained_model = \"tts_en_fastpitch\"\n",
                "    elif spectrogram_generator == \"mixertts\":\n",
                "        from nemo.collections.tts.models import MixerTTSModel\n",
                "        pretrained_model = \"tts_en_lj_mixertts\"\n",
                "    elif spectrogram_generator == \"mixerttsx\":\n",
                "        from nemo.collections.tts.models import MixerTTSModel\n",
                "        pretrained_model = \"tts_en_lj_mixerttsx\"\n",
                "    else:\n",
                "        raise NotImplementedError\n",
                "        \n",
                "    model = from_pretrained_call(pretrained_model, override_config_path=override_conf)\n",
                "    \n",
                "    return model\n",
                "\n",
                "\n",
                "def load_vocoder_model():\n",
                "    TwoStagesModel = False\n",
                "    strict=True\n",
                "    \n",
                "    if audio_generator == \"waveglow\":\n",
                "        from nemo.collections.tts.models import WaveGlowModel\n",
                "        pretrained_model = \"tts_waveglow\"\n",
                "        strict=False\n",
                "    elif audio_generator == \"hifigan\":\n",
                "        from nemo.collections.tts.models import HifiGanModel\n",
                "        spectrogram_generator2ft_hifigan = {\n",
                "            \"mixertts\": \"tts_en_lj_hifigan_ft_mixertts\",\n",
                "            \"mixerttsx\": \"tts_en_lj_hifigan_ft_mixerttsx\"\n",
                "        }\n",
                "        pretrained_model = spectrogram_generator2ft_hifigan.get(spectrogram_generator, \"tts_en_hifigan\")\n",
                "    elif audio_generator == \"univnet\":\n",
                "        from nemo.collections.tts.models import UnivNetModel\n",
                "        pretrained_model = \"tts_en_lj_univnet\"\n",
                "    elif audio_generator == \"griffin-lim\":\n",
                "        from nemo.collections.tts.models import TwoStagesModel\n",
                "        cfg = {'linvocoder':  {'_target_': 'nemo.collections.tts.models.two_stages.GriffinLimModel',\n",
                "                             'cfg': {'n_iters': 64, 'n_fft': 1024, 'l_hop': 256}},\n",
                "               'mel2spec': {'_target_': 'nemo.collections.tts.models.two_stages.MelPsuedoInverseModel',\n",
                "                           'cfg': {'sampling_rate': 22050, 'n_fft': 1024, \n",
                "                                   'mel_fmin': 0, 'mel_fmax': 8000, 'mel_freq': 80}}}\n",
                "        model = TwoStagesModel(cfg)            \n",
                "        TwoStagesModel = True\n",
                "    else:\n",
                "        raise NotImplementedError\n",
                "\n",
                "    if not TwoStagesModel:\n",
                "        model = Vocoder.from_pretrained(pretrained_model, strict=strict)\n",
                "        \n",
                "    return model\n",
                "\n",
                "spec_gen = load_spectrogram_model().eval().cuda()\n",
                "vocoder = load_vocoder_model().eval().cuda()"
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "## Inference\n",
                "\n",
                "Now that we have downloaded the model checkpoints and loaded them into memory. Let's define a short infer helper function that takes a string, and our models to produce speech.\n",
                "\n",
                "Notice that the NeMo TTS model interface is fairly simple and standardized across all models.\n",
                "\n",
                "End-to-end models have two helper functions:\n",
                "- parse(): Accepts raw python strings and returns a torch.tensor that represents tokenized text\n",
                "- convert_text_to_waveform(): Accepts a batch of tokenized text and returns a torch.tensor that represents a batch of raw audio\n",
                "\n",
                "Mel Spectrogram generators have two helper functions:\n",
                "\n",
                "- parse(): Accepts raw python strings and returns a torch.tensor that represents tokenized text\n",
                "- generate_spectrogram(): Accepts a batch of tokenized text and returns a torch.tensor that represents a batch of spectrograms\n",
                "\n",
                "Vocoder have just one helper function:\n",
                "\n",
                "- convert_spectrogram_to_audio(): Accepts a batch of spectrograms and returns a torch.tensor that represents a batch of raw audio"
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {},
            "outputs": [],
            "source": [
                "def infer(spec_gen_model, vocoder_model, str_input):\n",
                "    parser_model = spec_gen_model\n",
                "    with torch.no_grad():\n",
                "        parsed = parser_model.parse(str_input)\n",
                "        gen_spec_kwargs = {}\n",
                "        \n",
                "        if spectrogram_generator == \"mixerttsx\":\n",
                "            gen_spec_kwargs[\"raw_texts\"] = [str_input]\n",
                "        \n",
                "        spectrogram = spec_gen_model.generate_spectrogram(tokens=parsed, **gen_spec_kwargs)\n",
                "        audio = vocoder_model.convert_spectrogram_to_audio(spec=spectrogram)\n",
                "        \n",
                "        if audio_generator == \"hifigan\":\n",
                "            audio = vocoder_model._bias_denoise(audio, spectrogram).squeeze(1)\n",
                "    if spectrogram is not None:\n",
                "        if isinstance(spectrogram, torch.Tensor):\n",
                "            spectrogram = spectrogram.to('cpu').numpy()\n",
                "        if len(spectrogram.shape) == 3:\n",
                "            spectrogram = spectrogram[0]\n",
                "    if isinstance(audio, torch.Tensor):\n",
                "        audio = audio.to('cpu').numpy()\n",
                "    return spectrogram, audio"
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "Now that everything is set up, let's give an input that we want our models to speak"
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {},
            "outputs": [],
            "source": [
                "text_to_generate = input(\"Input what you want the model to say: \")\n",
                "spec, audio = infer(spec_gen, vocoder, text_to_generate)"
            ]
        },
        {
            "cell_type": "markdown",
            "metadata": {},
            "source": [
                "# Results\n",
                "\n",
                "After our model generates the audio, let's go ahead and play it. We can also visualize the spectrogram that was produced from the first stage model if a spectrogram generator was used."
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {},
            "outputs": [],
            "source": [
                "import IPython.display as ipd\n",
                "import numpy as np\n",
                "from PIL import Image\n",
                "from matplotlib.pyplot import imshow\n",
                "from matplotlib import pyplot as plt\n",
                "\n",
                "ipd.Audio(audio, rate=22050)"
            ]
        },
        {
            "cell_type": "code",
            "execution_count": null,
            "metadata": {},
            "outputs": [],
            "source": [
                "%matplotlib inline\n",
                "if spec is not None:\n",
                "    imshow(spec, origin=\"lower\")\n",
                "    plt.show()"
            ]
        }
    ],
    "metadata": {
        "interpreter": {
            "hash": "9750c3195c8cb9412c65c8ba8b36c6ba2b82b23ddd61f39d29e01b403b049680"
        },
        "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": "3.8.6"
        }
    },
    "nbformat": 4,
    "nbformat_minor": 4
}