def __init__(self, filelist_path, filter_length, hop_length, win_length,
sampling_rate, mel_fmin, mel_fmax, max_wav_value, p_arpabet,
cmudict_path, text_cleaners, speaker_ids=None,
use_attn_prior=False, attn_prior_threshold=1e-4,
randomize=True, seed=1234):
self.max_wav_value = max_wav_value
self.audiopaths_and_text = load_filepaths_and_text(filelist_path)
self.use_attn_prior = use_attn_prior
self.attn_prior_threshold = attn_prior_threshold
if speaker_ids is None or speaker_ids == '':
self.speaker_ids = self.create_speaker_lookup_table(self.audiopaths_and_text)
else:
self.speaker_ids = speaker_ids
self.stft = TacotronSTFT(filter_length=filter_length,
hop_length=hop_length,
win_length=win_length,
sampling_rate=sampling_rate,
mel_fmin=mel_fmin, mel_fmax=mel_fmax)
self.sampling_rate = sampling_rate
self.text_cleaners = text_cleaners
self.p_arpabet = p_arpabet
self.cmudict = cmudict.CMUDict(cmudict_path, keep_ambiguous=True)
if speaker_ids is None:
self.speaker_ids = self.create_speaker_lookup_table(self.audiopaths_and_text)
else:
self.speaker_ids = speaker_ids
random.seed(seed)
if randomize:
random.shuffle(self.audiopaths_and_text)
def __init__(self,
filelist_path,
filter_length,
hop_length,
win_length,
sampling_rate,
mel_fmin,
mel_fmax,
max_wav_value,
p_arpabet,
cmudict_path,
text_cleaners,
speaker_ids=None,
use_attn_prior=False,
attn_prior_threshold=1e-4,
prior_cache_path="",
betab_scaling_factor=1.0,
randomize=True,
keep_ambiguous=False,
seed=1234):
self.max_wav_value = max_wav_value
self.audiopaths_and_text = load_filepaths_and_text(filelist_path)
self.use_attn_prior = use_attn_prior
self.betab_scaling_factor = betab_scaling_factor
self.attn_prior_threshold = attn_prior_threshold
self.keep_ambiguous = keep_ambiguous
if speaker_ids is None or speaker_ids == '':
self.speaker_ids = self.create_speaker_lookup_table(
self.audiopaths_and_text)
else:
self.speaker_ids = speaker_ids
self.stft = TacotronSTFT(filter_length=filter_length,
hop_length=hop_length,
win_length=win_length,
sampling_rate=sampling_rate,
mel_fmin=mel_fmin,
mel_fmax=mel_fmax)
self.sampling_rate = sampling_rate
self.text_cleaners = text_cleaners
self.p_arpabet = p_arpabet
self.cmudict = cmudict.CMUDict(cmudict_path,
keep_ambiguous=keep_ambiguous)
if speaker_ids is None:
self.speaker_ids = self.create_speaker_lookup_table(
self.audiopaths_and_text)
else:
self.speaker_ids = speaker_ids
# caching makes sense for p_phoneme=1.0
# for other values, everytime text lengths will change
self.prior_cache_path = prior_cache_path
self.caching_enabled = False
if (self.prior_cache_path is not None and self.prior_cache_path != ""
and p_arpabet == 1.0):
self.caching_enabled = True
# make sure caching path exists
if (self.caching_enabled
and not os.path.exists(self.prior_cache_path)):
os.makedirs(self.prior_cache_path)
random.seed(seed)
if randomize:
random.shuffle(self.audiopaths_and_text)