Merge pull request #1 from mozilla/dev

Sync with Mozilla TTS dev branch

TTS/bin/train_encoder.py

@@ -13,10 +13,9 @@ from TTS.speaker_encoder.dataset import MyDataset
 from TTS.speaker_encoder.losses import AngleProtoLoss, GE2ELoss
 from TTS.speaker_encoder.model import SpeakerEncoder
 from TTS.speaker_encoder.utils.generic_utils import \
-    check_config_speaker_encoder
+    check_config_speaker_encoder, save_best_model
 from TTS.speaker_encoder.utils.visual import plot_embeddings
 from TTS.tts.datasets.preprocess import load_meta_data
-from TTS.tts.utils.io import save_best_model
 from TTS.utils.audio import AudioProcessor
 from TTS.utils.generic_utils import (count_parameters,
                                      create_experiment_folder, get_git_branch,

TTS/bin/train_glow_tts.py

@@ -57,6 +57,7 @@ def setup_loader(ap, r, is_val=False, verbose=False):
             use_phonemes=c.use_phonemes,
             phoneme_language=c.phoneme_language,
             enable_eos_bos=c.enable_eos_bos_chars,
+            use_noise_augment=not is_val,
             verbose=verbose,
             speaker_mapping=speaker_mapping if c.use_speaker_embedding and c.use_external_speaker_embedding_file else None)
 
@@ -279,7 +280,12 @@ def train(data_loader, model, criterion, optimizer, scheduler,
                 # Diagnostic visualizations
                 # direct pass on model for spec predictions
                 target_speaker = None if speaker_c is None else speaker_c[:1]
-                spec_pred, *_ = model.inference(text_input[:1], text_lengths[:1], g=target_speaker)
+
+                if hasattr(model, 'module'):
+                    spec_pred, *_ = model.module.inference(text_input[:1], text_lengths[:1], g=target_speaker)
+                else:
+                    spec_pred, *_ = model.inference(text_input[:1], text_lengths[:1], g=target_speaker)
+
                 spec_pred = spec_pred.permute(0, 2, 1)
                 gt_spec = mel_input.permute(0, 2, 1)
                 const_spec = spec_pred[0].data.cpu().numpy()

TTS/tts/datasets/TTSDataset.py

@@ -30,6 +30,7 @@ class MyDataset(Dataset):
                  phoneme_language="en-us",
                  enable_eos_bos=False,
                  speaker_mapping=None,
+                 use_noise_augment=False,
                  verbose=False):
         """
         Args:
@@ -48,6 +49,7 @@ class MyDataset(Dataset):
             phoneme_language (str): one the languages from
                 https://github.com/bootphon/phonemizer#languages
             enable_eos_bos (bool): enable end of sentence and beginning of sentences characters.
+            use_noise_augment (bool): enable adding random noise to wav for augmentation.
             verbose (bool): print diagnostic information.
         """
         self.batch_group_size = batch_group_size
@@ -66,6 +68,7 @@ class MyDataset(Dataset):
         self.phoneme_language = phoneme_language
         self.enable_eos_bos = enable_eos_bos
         self.speaker_mapping = speaker_mapping
+        self.use_noise_augment = use_noise_augment
         self.verbose = verbose
         self.input_seq_computed = False
         if use_phonemes and not os.path.isdir(phoneme_cache_path):
@@ -134,6 +137,10 @@ class MyDataset(Dataset):
 
         wav = np.asarray(self.load_wav(wav_file), dtype=np.float32)
 
+        # apply noise for augmentation
+        if self.use_noise_augment:
+            wav = wav + (1.0 / 32768.0) * np.random.rand(*wav.shape)
+
         if not self.input_seq_computed:
             if self.use_phonemes:
                 text = self._load_or_generate_phoneme_sequence(wav_file, text, self.phoneme_cache_path, self.enable_eos_bos, self.cleaners, self.phoneme_language, self.tp, self.add_blank)

TTS/tts/utils/synthesis.py

@@ -62,7 +62,11 @@ def run_model_torch(model, inputs, CONFIG, truncated, speaker_id=None, style_mel
                     inputs, speaker_ids=speaker_id, speaker_embeddings=speaker_embeddings)
     elif 'glow' in CONFIG.model.lower():
         inputs_lengths = torch.tensor(inputs.shape[1:2]).to(inputs.device)  # pylint: disable=not-callable
-        postnet_output, _, _, _, alignments, _, _ = model.inference(inputs, inputs_lengths, g=speaker_id if speaker_id else speaker_embeddings)
+        if hasattr(model, 'module'):
+            # distributed model
+            postnet_output, _, _, _, alignments, _, _ = model.module.inference(inputs, inputs_lengths, g=speaker_id if speaker_id is not None else speaker_embeddings)
+        else:
+            postnet_output, _, _, _, alignments, _, _ = model.inference(inputs, inputs_lengths, g=speaker_id if speaker_id is not None else speaker_embeddings)
         postnet_output = postnet_output.permute(0, 2, 1)
         # these only belong to tacotron models.
         decoder_output = None

TTS/tts/utils/text/cleaners.py

@@ -14,6 +14,7 @@ import re
 from unidecode import unidecode
 from .number_norm import normalize_numbers
 from .abbreviations import abbreviations_en, abbreviations_fr
+from .time import expand_time_english
 
 # Regular expression matching whitespace:
 _whitespace_re = re.compile(r'\s+')
@@ -95,6 +96,7 @@ def english_cleaners(text):
     '''Pipeline for English text, including number and abbreviation expansion.'''
     text = convert_to_ascii(text)
     text = lowercase(text)
+    text = expand_time_english(text)
     text = expand_numbers(text)
     text = expand_abbreviations(text)
     text = replace_symbols(text)
@@ -122,8 +124,8 @@ def portuguese_cleaners(text):
 
 def phoneme_cleaners(text):
     '''Pipeline for phonemes mode, including number and abbreviation expansion.'''
-    text = convert_to_ascii(text)
     text = expand_numbers(text)
+    text = convert_to_ascii(text)
     text = expand_abbreviations(text)
     text = replace_symbols(text)
     text = remove_aux_symbols(text)

TTS/tts/utils/text/number_norm.py

@@ -2,14 +2,14 @@
 
 import inflect
 import re
+from typing import Dict
 
 _inflect = inflect.engine()
 _comma_number_re = re.compile(r'([0-9][0-9\,]+[0-9])')
 _decimal_number_re = re.compile(r'([0-9]+\.[0-9]+)')
-_pounds_re = re.compile(r'£([0-9\,]*[0-9]+)')
-_dollars_re = re.compile(r'\$([0-9\.\,]*[0-9]+)')
+_currency_re = re.compile(r'(£|\$|¥)([0-9\,\.]*[0-9]+)')
 _ordinal_re = re.compile(r'[0-9]+(st|nd|rd|th)')
-_number_re = re.compile(r'[0-9]+')
+_number_re = re.compile(r'-?[0-9]+')
 
 
 def _remove_commas(m):
@@ -20,24 +20,54 @@ def _expand_decimal_point(m):
     return m.group(1).replace('.', ' point ')
 
 
-def _expand_dollars(m):
-    match = m.group(1)
-    parts = match.split('.')
+def __expand_currency(value: str, inflection: Dict[float, str]) -> str:
+    parts = value.replace(",", "").split('.')
     if len(parts) > 2:
-        return match + ' dollars'  # Unexpected format
-    dollars = int(parts[0]) if parts[0] else 0
-    cents = int(parts[1]) if len(parts) > 1 and parts[1] else 0
-    if dollars and cents:
-        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
-        cent_unit = 'cent' if cents == 1 else 'cents'
-        return '%s %s, %s %s' % (dollars, dollar_unit, cents, cent_unit)
-    if dollars:
-        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
-        return '%s %s' % (dollars, dollar_unit)
-    if cents:
-        cent_unit = 'cent' if cents == 1 else 'cents'
-        return '%s %s' % (cents, cent_unit)
-    return 'zero dollars'
+        return f"{value} {inflection[2]}"  # Unexpected format
+    text = []
+    integer = int(parts[0]) if parts[0] else 0
+    if integer > 0:
+        integer_unit = inflection.get(integer, inflection[2])
+        text.append(f"{integer} {integer_unit}")
+    fraction = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    if fraction > 0:
+        fraction_unit = inflection.get(fraction/100, inflection[0.02])
+        text.append(f"{fraction} {fraction_unit}")
+    if len(text) == 0:
+        return f"zero {inflection[2]}"
+    return " ".join(text)
+
+
+def _expand_currency(m: "re.Match") -> str:
+    currencies = {
+        "$": {
+            0.01: "cent",
+            0.02: "cents",
+            1: "dollar",
+            2: "dollars",
+        },
+        "€": {
+            0.01: "cent",
+            0.02: "cents",
+            1: "euro",
+            2: "euros",
+        },
+        "£": {
+            0.01: "penny",
+            0.02: "pence",
+            1: "pound sterling",
+            2: "pounds sterling",
+        },
+        "¥": {
+            # TODO rin
+            0.02: "sen",
+            2: "yen",
+        }
+    }
+    unit = m.group(1)
+    currency = currencies[unit]
+    value = m.group(2)
+    return __expand_currency(value, currency)
 
 
 def _expand_ordinal(m):
@@ -62,8 +92,7 @@ def _expand_number(m):
 
 def normalize_numbers(text):
     text = re.sub(_comma_number_re, _remove_commas, text)
-    text = re.sub(_pounds_re, r'\1 pounds', text)
-    text = re.sub(_dollars_re, _expand_dollars, text)
+    text = re.sub(_currency_re, _expand_currency, text)
     text = re.sub(_decimal_number_re, _expand_decimal_point, text)
     text = re.sub(_ordinal_re, _expand_ordinal, text)
     text = re.sub(_number_re, _expand_number, text)

TTS/tts/utils/text/time.py

@@ -0,0 +1,44 @@
+import re
+import inflect
+
+_inflect = inflect.engine()
+
+_time_re = re.compile(r"""\b
+                          ((0?[0-9])|(1[0-1])|(1[2-9])|(2[0-3]))  # hours
+                          :
+                          ([0-5][0-9])                            # minutes
+                          \s*(a\\.m\\.|am|pm|p\\.m\\.|a\\.m|p\\.m)? # am/pm
+                          \b""",
+                      re.IGNORECASE | re.X)
+
+
+def _expand_num(n: int) -> str:
+    return _inflect.number_to_words(n)
+
+
+def _expand_time_english(match: "re.Match") -> str:
+    hour = int(match.group(1))
+    past_noon = hour >= 12
+    time = []
+    if hour > 12:
+        hour -= 12
+    elif hour == 0:
+        hour = 12
+        past_noon = True
+    time.append(_expand_num(hour))
+
+    minute = int(match.group(6))
+    if minute > 0:
+        if minute < 10:
+            time.append("oh")
+        time.append(_expand_num(minute))
+    am_pm = match.group(7)
+    if am_pm is None:
+        time.append("p m" if past_noon else "a m")
+    else:
+        time.extend(list(am_pm.replace(".", "")))
+    return " ".join(time)
+
+
+def expand_time_english(text: str) -> str:
+    return re.sub(_time_re, _expand_time_english, text)

tests/test_text_cleaners.py

@@ -0,0 +1,21 @@
+#!/usr/bin/env python3
+
+from TTS.tts.utils.text.cleaners import english_cleaners, phoneme_cleaners
+
+
+def test_time() -> None:
+    assert english_cleaners("It's 11:00") == "it's eleven a m"
+    assert english_cleaners("It's 9:01") == "it's nine oh one a m"
+    assert english_cleaners("It's 16:00") == "it's four p m"
+    assert english_cleaners("It's 00:00 am") == "it's twelve a m"
+
+
+def test_currency() -> None:
+    assert phoneme_cleaners("It's $10.50") == "It's ten dollars fifty cents"
+    assert phoneme_cleaners("£1.1") == "one pound sterling one penny"
+    assert phoneme_cleaners("¥1") == "one yen"
+
+
+def test_expand_numbers() -> None:
+    assert "minus one" == phoneme_cleaners("-1")
+    assert "one" == phoneme_cleaners("1")

tests/test_text_processing.py

@@ -171,4 +171,4 @@ def test_text2phone():
     gt = "ɹ|iː|s|ə|n|t| |ɹ|ɪ|s|ɜː|tʃ| |æ|t| |h|ɑːɹ|v|ɚ|d| |h|ɐ|z| |ʃ|oʊ|n| |m|ɛ|d|ᵻ|t|eɪ|ɾ|ɪ|ŋ| |f|ɔː|ɹ| |æ|z| |l|ɪ|ɾ|əl| |æ|z| |eɪ|t| |w|iː|k|s| |k|æ|n| |æ|k|tʃ|uː|əl|i| |ɪ|n|k|ɹ|iː|s|,| |ð|ə| |ɡ|ɹ|eɪ| |m|æ|ɾ|ɚ|ɹ| |ɪ|n|ð|ə| |p|ɑːɹ|t|s| |ʌ|v|ð|ə| |b|ɹ|eɪ|n| |ɹ|ɪ|s|p|ɑː|n|s|ə|b|əl| |f|ɔː|ɹ| |ɪ|m|oʊ|ʃ|ə|n|əl| |ɹ|ɛ|ɡ|j|uː|l|eɪ|ʃ|ə|n| |æ|n|d| |l|ɜː|n|ɪ|ŋ|!"
     lang = "en-us"
     ph = text2phone(text, lang)
-    assert gt == ph, f"\n{phonemes} \n vs \n{gt}"
+    assert gt == ph