malarinv
/
tacotron2
mirror of https://github.com/malarinv/tacotron2
1
0
Fork 0
Code Issues Releases Wiki Activity
tacotron2/inference.ipynb

267 lines
330 KiB
Plaintext
Raw Normal View History

inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Tacotron 2 inference code \n",
"Edit the variables **checkpoint_path** and **text** to match yours and run the entire code to generate plots of mel outputs, alignments and audio synthesis from the generated mel-spectrogram using Griffin-Lim."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Import libraries and setup matplotlib"
]
},
{
"cell_type": "code",
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"execution_count": 1,
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"metadata": {},
"outputs": [],
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"source": [
"import matplotlib\n",
inference.ipynb: setting relative model paths
2018-11-27 15:45:09 +00:00
"%matplotlib inline\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"import matplotlib.pylab as plt\n",
"\n",
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"import IPython.display as ipd\n",
"\n",
inference.ipynb: setting relative model paths
2018-11-27 15:45:09 +00:00
"import sys\n",
"sys.path.append('waveglow/')\n",
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"import numpy as np\n",
"import torch\n",
"\n",
"from hparams import create_hparams\n",
"from model import Tacotron2\n",
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
"from layers import TacotronSTFT, STFT\n",
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"from audio_processing import griffin_lim\n",
"from train import load_model\n",
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
"from text import text_to_sequence\n",
"from denoiser import Denoiser"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
},
{
"cell_type": "code",
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"execution_count": 2,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {},
"outputs": [],
"source": [
"def plot_data(data, figsize=(16, 4)):\n",
" fig, axes = plt.subplots(1, len(data), figsize=figsize)\n",
" for i in range(len(data)):\n",
" axes[i].imshow(data[i], aspect='auto', origin='bottom', \n",
" interpolation='none')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Setup hparams"
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 3,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {},
inference.ipynb: cleanup
2018-11-27 20:04:36 +00:00
"outputs": [],
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"source": [
inference.ipynb: cleanup
2018-11-27 20:04:36 +00:00
"hparams = create_hparams()\n",
"hparams.sampling_rate = 22050"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Load model from checkpoint"
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 4,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {},
"outputs": [],
"source": [
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"checkpoint_path = \"tacotron2_statedict.pt\"\n",
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"model = load_model(hparams)\n",
inference.ipynb: cleanup
2018-11-27 20:04:36 +00:00
"model.load_state_dict(torch.load(checkpoint_path)['state_dict'])\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"_ = model.cuda().eval().half()"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
},
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
{
"cell_type": "markdown",
"metadata": {},
"source": [
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
"#### Load WaveGlow for mel2audio synthesis and denoiser"
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 5,
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"metadata": {},
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"outputs": [
{
"name": "stderr",
"output_type": "stream",
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
"text": [ ]
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
}
],
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"source": [
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"waveglow_path = 'waveglow_256channels.pt'\n",
inference.ipynb: changing waverglow inference fo fp16
2018-12-06 06:14:35 +00:00
"waveglow = torch.load(waveglow_path)['model']\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"waveglow.cuda().eval().half()\n",
"for k in waveglow.convinv:\n",
" k.float()\n",
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
"denoiser = Denoiser(waveglow)"
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
]
},
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Prepare text input"
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 6,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {},
"outputs": [],
"source": [
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"text = \"Waveglow is really awesome!\"\n",
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"sequence = np.array(text_to_sequence(text, ['english_cleaners']))[None, :]\n",
"sequence = torch.autograd.Variable(\n",
" torch.from_numpy(sequence)).cuda().long()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Decode text input and plot results"
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 7,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {
"scrolled": true
},
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"outputs": [
{
"data": {
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"image/png": "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
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"text/plain": [
"<Figure size 1152x288 with 3 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"source": [
"mel_outputs, mel_outputs_postnet, _, alignments = model.inference(sequence)\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"plot_data((mel_outputs.float().data.cpu().numpy()[0],\n",
" mel_outputs_postnet.float().data.cpu().numpy()[0],\n",
" alignments.float().data.cpu().numpy()[0].T))"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"#### Synthesize audio from spectrogram using WaveGlow"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
},
{
"cell_type": "code",
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"execution_count": 8,
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"metadata": {},
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
"outputs": [
{
"data": {
"text/html": [
"\n",
" <audio controls=\"controls\" >\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
" <source src=\"data:audio/wav;base64,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
inference.ipynb: updating inference file with relative paths
2018-11-27 16:04:04 +00:00
" Your browser does not support the audio element.\n",
" </audio>\n",
" "
],
"text/plain": [
"<IPython.lib.display.Audio object>"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
}
],
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
"source": [
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"with torch.no_grad():\n",
inference.ipynb: reverting fp16 inference for now
2018-12-09 05:26:01 +00:00
" audio = waveglow.infer(mel_outputs_postnet, sigma=0.666)\n",
inference.ipynb: import taco2model to be public
2018-11-26 06:33:16 +00:00
"ipd.Audio(audio[0].data.cpu().numpy(), rate=hparams.sampling_rate)"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
]
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### (Optional) Remove WaveGlow bias"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" <audio controls=\"controls\" >\n",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
" <source src=\"data:audio/wav;base64,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
inference.ipynb: adding code to remove waveglows bias
2019-03-15 23:54:54 +00:00
" Your browser does not support the audio element.\n",
" </audio>\n",
" "
],
"text/plain": [
"<IPython.lib.display.Audio object>"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"audio_denoised = denoiser(audio, strength=0.01)[:, 0]\n",
"ipd.Audio(audio_denoised.cpu().numpy(), rate=hparams.sampling_rate) "
]
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"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",
inference.ipynb: adding fp16 inference
2019-04-03 21:41:11 +00:00
"version": "3.6.8"
inference.ipynb: adding notebook for griffin-lim inference
2018-05-03 22:16:32 +00:00
}
},
"nbformat": 4,
"nbformat_minor": 2
}