{
    "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
}