mirror of https://github.com/coqui-ai/TTS.git
fix invalid json (#1599)
This commit is contained in:
Ryan Le-Nguyen
GitHub
parent
a790df4e94
commit
b6bd74a9a9
|
|
@ -1,191 +1,188 @@
|
||||||
{
|
{
|
||||||
"cells": [
|
"cells": [{
|
||||||
{
|
"cell_type": "markdown",
|
||||||
"cell_type": "markdown",
|
"metadata": {
|
||||||
"metadata": {
|
"Collapsed": "false"
|
||||||
"Collapsed": "false"
|
},
|
||||||
},
|
"source": [
|
||||||
"source": [
|
"This notebook is to test attention performance of a TTS model on a list of sentences taken from DeepVoice paper.\n",
|
||||||
"This notebook is to test attention performance of a TTS model on a list of sentences taken from DeepVoice paper.\n",
|
"### Features of this notebook\n",
|
||||||
"### Features of this notebook\n",
|
"- You can see visually how your model performs on each sentence and try to dicern common problems.\n",
|
||||||
"- You can see visually how your model performs on each sentence and try to dicern common problems.\n",
|
"- At the end, final attention score would be printed showing the ultimate performace of your model. You can use this value to perform model selection.\n",
|
||||||
"- At the end, final attention score would be printed showing the ultimate performace of your model. You can use this value to perform model selection.\n",
|
"- You can change the list of sentences byt providing a different sentence file."
|
||||||
"- You can change the list of sentences byt providing a different sentence file."
|
]
|
||||||
]
|
},
|
||||||
},
|
{
|
||||||
{
|
"cell_type": "code",
|
||||||
"cell_type": "code",
|
"execution_count": null,
|
||||||
"execution_count": null,
|
"metadata": {
|
||||||
"metadata": {
|
"Collapsed": "false",
|
||||||
"Collapsed": "false",
|
"scrolled": true
|
||||||
"scrolled": true
|
},
|
||||||
},
|
"outputs": [],
|
||||||
"outputs": [],
|
"source": [
|
||||||
"source": [
|
"%load_ext autoreload\n",
|
||||||
"%load_ext autoreload\n",
|
"%autoreload 2\n",
|
||||||
"%autoreload 2\n",
|
"import os, sys\n",
|
||||||
"import os, sys\n",
|
"import torch \n",
|
||||||
"import torch \n",
|
"import time\n",
|
||||||
"import time\n",
|
"import numpy as np\n",
|
||||||
"import numpy as np\n",
|
"from matplotlib import pylab as plt\n",
|
||||||
"from matplotlib import pylab as plt\n",
|
"\n",
|
||||||
"\n",
|
"%pylab inline\n",
|
||||||
"%pylab inline\n",
|
"plt.rcParams[\"figure.figsize\"] = (16,5)\n",
|
||||||
"plt.rcParams[\"figure.figsize\"] = (16,5)\n",
|
"\n",
|
||||||
"\n",
|
"import librosa\n",
|
||||||
"import librosa\n",
|
"import librosa.display\n",
|
||||||
"import librosa.display\n",
|
"\n",
|
||||||
"\n",
|
"from TTS.tts.layers import *\n",
|
||||||
"from TTS.tts.layers import *\n",
|
"from TTS.utils.audio import AudioProcessor\n",
|
||||||
"from TTS.utils.audio import AudioProcessor
|
"from TTS.tts.utils.generic_utils import setup_model\n",
|
||||||
|
"from TTS.tts.utils.io import load_config\n",
|
||||||
\n",
|
"from TTS.tts.utils.text import text_to_sequence\n",
|
||||||
"from TTS.tts.utils.generic_utils import setup_model\n",
|
"from TTS.tts.utils.synthesis import synthesis\n",
|
||||||
"from TTS.tts.utils.io import load_config\n",
|
"from TTS.tts.utils.visual import plot_alignment\n",
|
||||||
"from TTS.tts.utils.text import text_to_sequence\n",
|
"from TTS.tts.utils.measures import alignment_diagonal_score\n",
|
||||||
"from TTS.tts.utils.synthesis import synthesis\n",
|
"\n",
|
||||||
"from TTS.tts.utils.visual import plot_alignment\n",
|
"import IPython\n",
|
||||||
"from TTS.tts.utils.measures import alignment_diagonal_score\n",
|
"from IPython.display import Audio\n",
|
||||||
"\n",
|
"\n",
|
||||||
"import IPython\n",
|
"os.environ['CUDA_VISIBLE_DEVICES']='1'\n",
|
||||||
"from IPython.display import Audio\n",
|
"\n",
|
||||||
"\n",
|
"def tts(model, text, CONFIG, use_cuda, ap):\n",
|
||||||
"os.environ['CUDA_VISIBLE_DEVICES']='1'\n",
|
" t_1 = time.time()\n",
|
||||||
"\n",
|
" # run the model\n",
|
||||||
"def tts(model, text, CONFIG, use_cuda, ap):\n",
|
" waveform, alignment, mel_spec, mel_postnet_spec, stop_tokens, inputs = synthesis(model, text, CONFIG, use_cuda, ap, speaker_id, None, False, CONFIG.enable_eos_bos_chars, True)\n",
|
||||||
" t_1 = time.time()\n",
|
" if CONFIG.model == \"Tacotron\" and not use_gl:\n",
|
||||||
" # run the model\n",
|
" mel_postnet_spec = ap.out_linear_to_mel(mel_postnet_spec.T).T\n",
|
||||||
" waveform, alignment, mel_spec, mel_postnet_spec, stop_tokens, inputs = synthesis(model, text, CONFIG, use_cuda, ap, speaker_id, None, False, CONFIG.enable_eos_bos_chars, True)\n",
|
" # plotting\n",
|
||||||
" if CONFIG.model == \"Tacotron\" and not use_gl:\n",
|
" attn_score = alignment_diagonal_score(torch.FloatTensor(alignment).unsqueeze(0))\n",
|
||||||
" mel_postnet_spec = ap.out_linear_to_mel(mel_postnet_spec.T).T\n",
|
" print(f\" > {text}\")\n",
|
||||||
" # plotting\n",
|
" IPython.display.display(IPython.display.Audio(waveform, rate=ap.sample_rate))\n",
|
||||||
" attn_score = alignment_diagonal_score(torch.FloatTensor(alignment).unsqueeze(0))\n",
|
" fig = plot_alignment(alignment, fig_size=(8, 5))\n",
|
||||||
" print(f\" > {text}\")\n",
|
" IPython.display.display(fig)\n",
|
||||||
" IPython.display.display(IPython.display.Audio(waveform, rate=ap.sample_rate))\n",
|
" #saving results\n",
|
||||||
" fig = plot_alignment(alignment, fig_size=(8, 5))\n",
|
" os.makedirs(OUT_FOLDER, exist_ok=True)\n",
|
||||||
" IPython.display.display(fig)\n",
|
" file_name = text[:200].replace(\" \", \"_\").replace(\".\",\"\") + \".wav\"\n",
|
||||||
" #saving results\n",
|
" out_path = os.path.join(OUT_FOLDER, file_name)\n",
|
||||||
" os.makedirs(OUT_FOLDER, exist_ok=True)\n",
|
" ap.save_wav(waveform, out_path)\n",
|
||||||
" file_name = text[:200].replace(\" \", \"_\").replace(\".\",\"\") + \".wav\"\n",
|
" return attn_score\n",
|
||||||
" out_path = os.path.join(OUT_FOLDER, file_name)\n",
|
"\n",
|
||||||
" ap.save_wav(waveform, out_path)\n",
|
"# Set constants\n",
|
||||||
" return attn_score\n",
|
"ROOT_PATH = '/home/erogol/Models/LJSpeech/ljspeech-May-20-2020_12+29PM-1835628/'\n",
|
||||||
"\n",
|
"MODEL_PATH = ROOT_PATH + '/best_model.pth'\n",
|
||||||
"# Set constants\n",
|
"CONFIG_PATH = ROOT_PATH + '/config.json'\n",
|
||||||
"ROOT_PATH = '/home/erogol/Models/LJSpeech/ljspeech-May-20-2020_12+29PM-1835628/'\n",
|
"OUT_FOLDER = './hard_sentences/'\n",
|
||||||
"MODEL_PATH = ROOT_PATH + '/best_model.pth'\n",
|
"CONFIG = load_config(CONFIG_PATH)\n",
|
||||||
"CONFIG_PATH = ROOT_PATH + '/config.json'\n",
|
"SENTENCES_PATH = 'sentences.txt'\n",
|
||||||
"OUT_FOLDER = './hard_sentences/'\n",
|
"use_cuda = True\n",
|
||||||
"CONFIG = load_config(CONFIG_PATH)\n",
|
"\n",
|
||||||
"SENTENCES_PATH = 'sentences.txt'\n",
|
"# Set some config fields manually for testing\n",
|
||||||
"use_cuda = True\n",
|
"# CONFIG.windowing = False\n",
|
||||||
"\n",
|
"# CONFIG.prenet_dropout = False\n",
|
||||||
"# Set some config fields manually for testing\n",
|
"# CONFIG.separate_stopnet = True\n",
|
||||||
"# CONFIG.windowing = False\n",
|
"CONFIG.use_forward_attn = False\n",
|
||||||
"# CONFIG.prenet_dropout = False\n",
|
"# CONFIG.forward_attn_mask = True\n",
|
||||||
"# CONFIG.separate_stopnet = True\n",
|
"# CONFIG.stopnet = True"
|
||||||
"CONFIG.use_forward_attn = False\n",
|
]
|
||||||
"# CONFIG.forward_attn_mask = True\n",
|
},
|
||||||
"# CONFIG.stopnet = True"
|
{
|
||||||
]
|
"cell_type": "code",
|
||||||
},
|
"execution_count": null,
|
||||||
{
|
"metadata": {
|
||||||
"cell_type": "code",
|
"Collapsed": "false"
|
||||||
"execution_count": null,
|
},
|
||||||
"metadata": {
|
"outputs": [],
|
||||||
"Collapsed": "false"
|
"source": [
|
||||||
},
|
"# LOAD TTS MODEL\n",
|
||||||
"outputs": [],
|
"from TTS.tts.utils.text.symbols import make_symbols, symbols, phonemes\n",
|
||||||
"source": [
|
"\n",
|
||||||
"# LOAD TTS MODEL\n",
|
"# multi speaker \n",
|
||||||
"from TTS.tts.utils.text.symbols import make_symbols, symbols, phonemes\n",
|
"if CONFIG.use_speaker_embedding:\n",
|
||||||
"\n",
|
" speakers = json.load(open(f\"{ROOT_PATH}/speakers.json\", 'r'))\n",
|
||||||
"# multi speaker \n",
|
" speakers_idx_to_id = {v: k for k, v in speakers.items()}\n",
|
||||||
"if CONFIG.use_speaker_embedding:\n",
|
"else:\n",
|
||||||
" speakers = json.load(open(f\"{ROOT_PATH}/speakers.json\", 'r'))\n",
|
" speakers = []\n",
|
||||||
" speakers_idx_to_id = {v: k for k, v in speakers.items()}\n",
|
" speaker_id = None\n",
|
||||||
"else:\n",
|
"\n",
|
||||||
" speakers = []\n",
|
"# if the vocabulary was passed, replace the default\n",
|
||||||
" speaker_id = None\n",
|
"if 'characters' in CONFIG.keys():\n",
|
||||||
"\n",
|
" symbols, phonemes = make_symbols(**CONFIG.characters)\n",
|
||||||
"# if the vocabulary was passed, replace the default\n",
|
"\n",
|
||||||
"if 'characters' in CONFIG.keys():\n",
|
"# load the model\n",
|
||||||
" symbols, phonemes = make_symbols(**CONFIG.characters)\n",
|
"num_chars = len(phonemes) if CONFIG.use_phonemes else len(symbols)\n",
|
||||||
"\n",
|
"model = setup_model(num_chars, len(speakers), CONFIG)\n",
|
||||||
"# load the model\n",
|
"\n",
|
||||||
"num_chars = len(phonemes) if CONFIG.use_phonemes else len(symbols)\n",
|
"# load the audio processor\n",
|
||||||
"model = setup_model(num_chars, len(speakers), CONFIG)\n",
|
"ap = AudioProcessor(**CONFIG.audio) \n",
|
||||||
"\n",
|
"\n",
|
||||||
"# load the audio processor\n",
|
"\n",
|
||||||
"ap = AudioProcessor(**CONFIG.audio) \n",
|
"# load model state\n",
|
||||||
"\n",
|
"if use_cuda:\n",
|
||||||
"\n",
|
" cp = torch.load(MODEL_PATH)\n",
|
||||||
"# load model state\n",
|
"else:\n",
|
||||||
"if use_cuda:\n",
|
" cp = torch.load(MODEL_PATH, map_location=lambda storage, loc: storage)\n",
|
||||||
" cp = torch.load(MODEL_PATH)\n",
|
"\n",
|
||||||
"else:\n",
|
"# load the model\n",
|
||||||
" cp = torch.load(MODEL_PATH, map_location=lambda storage, loc: storage)\n",
|
"model.load_state_dict(cp['model'])\n",
|
||||||
"\n",
|
"if use_cuda:\n",
|
||||||
"# load the model\n",
|
" model.cuda()\n",
|
||||||
"model.load_state_dict(cp['model'])\n",
|
"model.eval()\n",
|
||||||
"if use_cuda:\n",
|
"print(cp['step'])\n",
|
||||||
" model.cuda()\n",
|
"print(cp['r'])\n",
|
||||||
"model.eval()\n",
|
"\n",
|
||||||
"print(cp['step'])\n",
|
"# set model stepsize\n",
|
||||||
"print(cp['r'])\n",
|
"if 'r' in cp:\n",
|
||||||
"\n",
|
" model.decoder.set_r(cp['r'])"
|
||||||
"# set model stepsize\n",
|
]
|
||||||
"if 'r' in cp:\n",
|
},
|
||||||
" model.decoder.set_r(cp['r'])"
|
{
|
||||||
]
|
"cell_type": "code",
|
||||||
},
|
"execution_count": null,
|
||||||
{
|
"metadata": {
|
||||||
"cell_type": "code",
|
"Collapsed": "false"
|
||||||
"execution_count": null,
|
},
|
||||||
"metadata": {
|
"outputs": [],
|
||||||
"Collapsed": "false"
|
"source": [
|
||||||
},
|
"model.decoder.max_decoder_steps=3000\n",
|
||||||
"outputs": [],
|
"attn_scores = []\n",
|
||||||
"source": [
|
"with open(SENTENCES_PATH, 'r') as f:\n",
|
||||||
"model.decoder.max_decoder_steps=3000\n",
|
" for text in f:\n",
|
||||||
"attn_scores = []\n",
|
" attn_score = tts(model, text, CONFIG, use_cuda, ap)\n",
|
||||||
"with open(SENTENCES_PATH, 'r') as f:\n",
|
" attn_scores.append(attn_score)"
|
||||||
" for text in f:\n",
|
]
|
||||||
" attn_score = tts(model, text, CONFIG, use_cuda, ap)\n",
|
},
|
||||||
" attn_scores.append(attn_score)"
|
{
|
||||||
]
|
"cell_type": "code",
|
||||||
},
|
"execution_count": null,
|
||||||
{
|
"metadata": {
|
||||||
"cell_type": "code",
|
"Collapsed": "false"
|
||||||
"execution_count": null,
|
},
|
||||||
"metadata": {
|
"outputs": [],
|
||||||
"Collapsed": "false"
|
"source": [
|
||||||
},
|
"np.mean(attn_scores)"
|
||||||
"outputs": [],
|
]
|
||||||
"source": [
|
}
|
||||||
"np.mean(attn_scores)"
|
],
|
||||||
]
|
"metadata": {
|
||||||
}
|
"kernelspec": {
|
||||||
],
|
"display_name": "Python 3",
|
||||||
"metadata": {
|
"language": "python",
|
||||||
"kernelspec": {
|
"name": "python3"
|
||||||
"display_name": "Python 3",
|
},
|
||||||
"language": "python",
|
"language_info": {
|
||||||
"name": "python3"
|
"codemirror_mode": {
|
||||||
},
|
"name": "ipython",
|
||||||
"language_info": {
|
"version": 3
|
||||||
"codemirror_mode": {
|
},
|
||||||
"name": "ipython",
|
"file_extension": ".py",
|
||||||
"version": 3
|
"mimetype": "text/x-python",
|
||||||
},
|
"name": "python",
|
||||||
"file_extension": ".py",
|
"nbconvert_exporter": "python",
|
||||||
"mimetype": "text/x-python",
|
"pygments_lexer": "ipython3",
|
||||||
"name": "python",
|
"version": "3.8.5"
|
||||||
"nbconvert_exporter": "python",
|
}
|
||||||
"pygments_lexer": "ipython3",
|
},
|
||||||
"version": "3.8.5"
|
"nbformat": 4,
|
||||||
}
|
"nbformat_minor": 4
|
||||||
},
|
}
|
||||||
"nbformat": 4,
|
|
||||||
"nbformat_minor": 4
|
|
||||||
}
|
|
||||||
Loading…
Reference in New Issue