Merge branch 'dev' of github.com:mozilla/TTS into dev

.travis/script

@@ -11,5 +11,7 @@ fi
 
 if [[ "$TEST_SUITE" == "unittest" ]]; then
   # Run tests on all pushes
+  pushd tts_namespace
   python -m unittest
+  popd
 fi

datasets/TTSDataset.py

@@ -5,8 +5,8 @@ import torch
 import random
 from torch.utils.data import Dataset
 
-from utils.text import text_to_sequence, phoneme_to_sequence, pad_with_eos_bos
+from TTS.utils.text import text_to_sequence, phoneme_to_sequence, pad_with_eos_bos
-from utils.data import prepare_data, prepare_tensor, prepare_stop_target
+from TTS.utils.data import prepare_data, prepare_tensor, prepare_stop_target
 
 
 class MyDataset(Dataset):

distribute.py

@@ -9,7 +9,7 @@ import torch.distributed as dist
 from torch.utils.data.sampler import Sampler
 from torch.autograd import Variable
 from torch._utils import _flatten_dense_tensors, _unflatten_dense_tensors
-from utils.generic_utils import load_config, create_experiment_folder
+from TTS.utils.generic_utils import load_config, create_experiment_folder
 
 
 class DistributedSampler(Sampler):

layers/losses.py

@@ -1,6 +1,6 @@
 from torch import nn
 from torch.nn import functional
-from utils.generic_utils import sequence_mask
+from TTS.utils.generic_utils import sequence_mask
 
 
 class L1LossMasked(nn.Module):

models/tacotron.py

@@ -1,7 +1,7 @@
 # coding: utf-8
 from torch import nn
-from layers.tacotron import Encoder, Decoder, PostCBHG
+from TTS.layers.tacotron import Encoder, Decoder, PostCBHG
-from utils.generic_utils import sequence_mask
+from TTS.utils.generic_utils import sequence_mask
 
 
 class Tacotron(nn.Module):

models/tacotron2.py

@@ -1,7 +1,7 @@
 from math import sqrt
 from torch import nn
-from layers.tacotron2 import Encoder, Decoder, Postnet
+from TTS.layers.tacotron2 import Encoder, Decoder, Postnet
-from utils.generic_utils import sequence_mask
+from TTS.utils.generic_utils import sequence_mask
 
 
 # TODO: match function arguments with tacotron

models/tacotrongst.py

@@ -1,8 +1,8 @@
 # coding: utf-8
 from torch import nn
-from layers.tacotron import Encoder, Decoder, PostCBHG
+from TTS.layers.tacotron import Encoder, Decoder, PostCBHG
-from layers.gst_layers import GST
+from TTS.layers.gst_layers import GST
-from utils.generic_utils import sequence_mask
+from TTS.utils.generic_utils import sequence_mask
 
 
 class TacotronGST(nn.Module):

notebooks/Benchmark.ipynb

@@ -138,7 +138,7 @@
    "outputs": [],
    "source": [
     "# LOAD TTS MODEL\n",
-    "from utils.text.symbols import symbols, phonemes\n",
+    "from TTS.utils.text.symbols import symbols, phonemes\n",
     "\n",
     "# multi speaker \n",
     "if CONFIG.use_speaker_embedding:\n",

notebooks/ExtractTTSpectrogram.ipynb

@@ -105,10 +105,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from utils.text.symbols import symbols, phonemes\n",
+    "from TTS.utils.text.symbols import symbols, phonemes\n",
-    "from utils.generic_utils import sequence_mask\n",
+    "from TTS.utils.generic_utils import sequence_mask\n",
-    "from layers.losses import L1LossMasked\n",
+    "from TTS.layers.losses import L1LossMasked\n",
-    "from utils.text.symbols import symbols, phonemes\n",
+    "from TTS.utils.text.symbols import symbols, phonemes\n",
     "\n",
     "# load the model\n",
     "num_chars = len(phonemes) if C.use_phonemes else len(symbols)\n",

server/server.py

@@ -1,7 +1,7 @@
 #!flask/bin/python
 import argparse
 from synthesizer import Synthesizer
-from utils.generic_utils import load_config
+from TTS.utils.generic_utils import load_config
 from flask import Flask, request, render_template, send_file
 
 parser = argparse.ArgumentParser()

server/synthesizer.py

@@ -5,11 +5,11 @@ import numpy as np
 import torch
 import sys
 
-from utils.audio import AudioProcessor
+from TTS.utils.audio import AudioProcessor
-from utils.generic_utils import load_config, setup_model
+from TTS.utils.generic_utils import load_config, setup_model
-from utils.text import phonemes, symbols
+from TTS.utils.text import phonemes, symbols
-from utils.speakers import load_speaker_mapping
+from TTS.utils.speakers import load_speaker_mapping
-from utils.synthesis import *
+from TTS.utils.synthesis import *
 
 import re
 alphabets = r"([A-Za-z])"

setup.py

@@ -63,8 +63,8 @@ setup(
     url='https://github.com/mozilla/TTS',
     description='Text to Speech with Deep Learning',
     license='MPL-2.0',
-    package_dir={'TTS': '.'},
+    package_dir={'': 'tts_namespace'},
-    packages=['TTS'] + ['TTS.' + pkg for pkg in find_packages()],
+    packages=find_packages('tts_namespace'),
     project_urls={
               'Documentation': 'https://github.com/mozilla/TTS/wiki',
               'Tracker': 'https://github.com/mozilla/TTS/issues',

synthesize.py

@@ -4,10 +4,10 @@ import argparse
 import torch
 import string
 
-from utils.synthesis import synthesis
+from TTS.utils.synthesis import synthesis
-from utils.generic_utils import load_config, setup_model
+from TTS.utils.generic_utils import load_config, setup_model
-from utils.text.symbols import symbols, phonemes
+from TTS.utils.text.symbols import symbols, phonemes
-from utils.audio import AudioProcessor
+from TTS.utils.audio import AudioProcessor
 
 
 def tts(model,

tests/generic_utils_text.py

@@ -1,8 +1,8 @@
 import unittest
 import torch as T
 
-from utils.generic_utils import save_checkpoint, save_best_model
+from TTS.utils.generic_utils import save_checkpoint, save_best_model
-from layers.tacotron import Prenet
+from TTS.layers.tacotron import Prenet
 
 OUT_PATH = '/tmp/test.pth.tar'
 

tests/inputs/server_config.json

@@ -1,5 +1,5 @@
 {
-    "tts_path":"tests/outputs/",  // tts model root folder
+    "tts_path":"TTS/tests/outputs/",  // tts model root folder
     "tts_file":"checkpoint_10.pth.tar",     // tts checkpoint file
     "tts_config":"dummy_model_config.json",     // tts config.json file
     "tts_speakers": null,           // json file listing speaker ids. null if no speaker embedding.

tests/symbols_tests.py

@@ -1,6 +1,6 @@
 import unittest
 
-from utils.text import phonemes
+from TTS.utils.text import phonemes
 
 class SymbolsTest(unittest.TestCase):
     def test_uniqueness(self):  #pylint: disable=no-self-use

tests/test_audio.py

@@ -1,9 +1,9 @@
 import os
 import unittest
 
-from tests import get_tests_path, get_tests_input_path, get_tests_output_path
+from TTS.tests import get_tests_path, get_tests_input_path, get_tests_output_path
-from utils.audio import AudioProcessor
+from TTS.utils.audio import AudioProcessor
-from utils.generic_utils import load_config
+from TTS.utils.generic_utils import load_config
 
 TESTS_PATH = get_tests_path()
 OUT_PATH = os.path.join(get_tests_output_path(), "audio_tests")

tests/test_demo_server.py

@@ -3,10 +3,10 @@ import unittest
 
 import torch as T
 
-from server.synthesizer import Synthesizer
+from TTS.server.synthesizer import Synthesizer
-from tests import get_tests_input_path, get_tests_output_path, get_tests_path
+from TTS.tests import get_tests_input_path, get_tests_output_path
-from utils.text.symbols import phonemes, symbols
+from TTS.utils.text.symbols import phonemes, symbols
-from utils.generic_utils import load_config, save_checkpoint, setup_model
+from TTS.utils.generic_utils import load_config, save_checkpoint, setup_model
 
 
 class DemoServerTest(unittest.TestCase):

tests/test_layers.py

@@ -1,9 +1,9 @@
 import unittest
 import torch as T
 
-from layers.tacotron import Prenet, CBHG, Decoder, Encoder
+from TTS.layers.tacotron import Prenet, CBHG, Decoder, Encoder
-from layers.losses import L1LossMasked
+from TTS.layers.losses import L1LossMasked
-from utils.generic_utils import sequence_mask
+from TTS.utils.generic_utils import sequence_mask
 
 #pylint: disable=unused-variable
 

tests/test_loader.py

@@ -5,10 +5,10 @@ import torch
 import numpy as np
 
 from torch.utils.data import DataLoader
-from utils.generic_utils import load_config
+from TTS.utils.generic_utils import load_config
-from utils.audio import AudioProcessor
+from TTS.utils.audio import AudioProcessor
-from datasets import TTSDataset
+from TTS.datasets import TTSDataset
-from datasets.preprocess import ljspeech
+from TTS.datasets.preprocess import ljspeech
 
 #pylint: disable=unused-variable
 

tests/test_preprocessors.py

@@ -1,8 +1,8 @@
 import unittest
 import os
-from tests import get_tests_input_path
+from TTS.tests import get_tests_input_path
 
-from datasets.preprocess import common_voice
+from TTS.datasets.preprocess import common_voice
 
 
 class TestPreprocessors(unittest.TestCase):

tests/test_tacotron2_model.py

@@ -6,9 +6,9 @@ import numpy as np
 
 from torch import optim
 from torch import nn
-from utils.generic_utils import load_config
+from TTS.utils.generic_utils import load_config
-from layers.losses import MSELossMasked
+from TTS.layers.losses import MSELossMasked
-from models.tacotron2 import Tacotron2
+from TTS.models.tacotron2 import Tacotron2
 
 #pylint: disable=unused-variable
 

tests/test_tacotron_model.py

@@ -5,9 +5,9 @@ import unittest
 
 from torch import optim
 from torch import nn
-from utils.generic_utils import load_config
+from TTS.utils.generic_utils import load_config
-from layers.losses import L1LossMasked
+from TTS.layers.losses import L1LossMasked
-from models.tacotron import Tacotron
+from TTS.models.tacotron import Tacotron
 
 #pylint: disable=unused-variable
 

tests/test_text_processing.py

@@ -1,7 +1,7 @@
 import unittest
 import torch as T
 
-from utils.text import *
+from TTS.utils.text import *
 
 def test_phoneme_to_sequence():
     text = "Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase, the grey matter in the parts of the brain responsible for emotional regulation and learning!"

train.py

@@ -10,24 +10,24 @@ import torch.nn as nn
 from torch import optim
 from torch.utils.data import DataLoader
 
-from datasets.TTSDataset import MyDataset
+from TTS.datasets.TTSDataset import MyDataset
 from distribute import (DistributedSampler, apply_gradient_allreduce,
                         init_distributed, reduce_tensor)
-from layers.losses import L1LossMasked, MSELossMasked
+from TTS.layers.losses import L1LossMasked, MSELossMasked
-from utils.audio import AudioProcessor
+from TTS.utils.audio import AudioProcessor
-from utils.generic_utils import (NoamLR, check_update, count_parameters,
+from TTS.utils.generic_utils import (NoamLR, check_update, count_parameters,
-                                 create_experiment_folder, get_git_branch,
+                                     create_experiment_folder, get_git_branch,
-                                 load_config, remove_experiment_folder,
+                                     load_config, remove_experiment_folder,
-                                 save_best_model, save_checkpoint, weight_decay,
+                                     save_best_model, save_checkpoint, weight_decay,
-                                 set_init_dict, copy_config_file, setup_model,
+                                     set_init_dict, copy_config_file, setup_model,
-                                 split_dataset, gradual_training_scheduler)
+                                     split_dataset, gradual_training_scheduler)
-from utils.logger import Logger
+from TTS.utils.logger import Logger
-from utils.speakers import load_speaker_mapping, save_speaker_mapping, \
+from TTS.utils.speakers import load_speaker_mapping, save_speaker_mapping, \
     get_speakers
-from utils.synthesis import synthesis
+from TTS.utils.synthesis import synthesis
-from utils.text.symbols import phonemes, symbols
+from TTS.utils.text.symbols import phonemes, symbols
-from utils.visual import plot_alignment, plot_spectrogram
+from TTS.utils.visual import plot_alignment, plot_spectrogram
-from datasets.preprocess import get_preprocessor_by_name
+from TTS.datasets.preprocess import get_preprocessor_by_name
 
 torch.backends.cudnn.enabled = True
 torch.backends.cudnn.benchmark = False

tts_namespace/README.md

@@ -0,0 +1,29 @@
+This folder contains a symlink called TTS to the parent folder:
+
+    lrwxr-xr-x  TTS -> ..
+
+This is used to appease the distribute/setuptools gods. When the project was
+initially set up, the repository folder itself was considered a namespace, and
+development was done with `sys.path` hacks. This means if you tried to install
+TTS, `setup.py` would see the packages `models`, `utils`, `layers`... instead of
+ `TTS.models`, `TTS.utils`...
+
+Installing TTS would then pollute the package namespace with generic names like
+those above. In order to make things installable in both install and development
+modes (`pip install /path/to/TTS` and `pip install -e /path/to/TTS`), we needed
+to add an additional 'TTS' namespace to avoid this pollution. A virtual redirect
+using `packages_dir` in `setup.py` is not enough because it breaks the editable
+installation, which can only handle the simplest of `package_dir` redirects.
+
+Our solution is to use a symlink in order to add the extra `TTS` namespace. In
+`setup.py`, we only look for packages inside `tts_namespace` (this folder),
+which contains a symlink called TTS pointing to the repository root. The final
+result is that `setuptools.find_packages` will find `TTS.models`, `TTS.utils`...
+
+With this hack, `pip install -e` will then add a symlink to the `tts_namespace`
+in your `site-packages` folder, which works properly. It's important not to add
+anything else in this folder because it will pollute the package namespace when
+installing the project.
+
+This does not work if you check out your project on a filesystem that does not
+support symlinks.

tts_namespace/TTS

@@ -0,0 +1 @@
+..

utils/generic_utils.py

@@ -250,7 +250,7 @@ def set_init_dict(model_dict, checkpoint, c):
 
 def setup_model(num_chars, num_speakers, c):
     print(" > Using model: {}".format(c.model))
-    MyModel = importlib.import_module('models.' + c.model.lower())
+    MyModel = importlib.import_module('TTS.models.' + c.model.lower())
     MyModel = getattr(MyModel, c.model)
     if c.model.lower() in ["tacotron", "tacotrongst"]:
         model = MyModel(

utils/speakers.py

@@ -1,7 +1,7 @@
 import os
 import json
 
-from datasets.preprocess import get_preprocessor_by_name
+from TTS.datasets.preprocess import get_preprocessor_by_name
 
 
 def make_speakers_json_path(out_path):

utils/text/__init__.py

@@ -3,8 +3,8 @@
 import re
 import phonemizer
 from phonemizer.phonemize import phonemize
-from utils.text import cleaners
+from TTS.utils.text import cleaners
-from utils.text.symbols import symbols, phonemes, _phoneme_punctuations, _bos, \
+from TTS.utils.text.symbols import symbols, phonemes, _phoneme_punctuations, _bos, \
     _eos
 
 # Mappings from symbol to numeric ID and vice versa:

utils/visual.py

@@ -2,7 +2,7 @@ import librosa
 import matplotlib
 matplotlib.use('Agg')
 import matplotlib.pyplot as plt
-from utils.text import phoneme_to_sequence, sequence_to_phoneme
+from TTS.utils.text import phoneme_to_sequence, sequence_to_phoneme
 
 
 def plot_alignment(alignment, info=None):