go to /content/tacotron_pytorch/hparams.py and change it to this:
import tensorflow as tf
import types
Default hyperparameters:
hparams_dict = {
Comma-separated list of cleaners to run on text prior to training and eval. For non-English
# text, you may want to use "basic_cleaners" or "transliteration_cleaners" See TRAINING_DATA.md.
'cleaners': 'english_cleaners',
'use_cmudict': False, # Use CMUDict during training to learn pronunciation of ARPAbet phonemes
# Audio:
'num_mels': 80,
'num_freq': 1025,
'sample_rate': 20000,
'frame_length_ms': 50,
'frame_shift_ms': 12.5,
'preemphasis': 0.97,
'min_level_db': -100,
'ref_level_db': 20,
# Model:
# TODO: add more configurable hparams
'outputs_per_step': 5,
'padding_idx': None,
'use_memory_mask': False,
# Data loader
'pin_memory': True,
'num_workers': 2,
# Training:
'batch_size': 32,
'adam_beta1': 0.9,
'adam_beta2': 0.999,
'initial_learning_rate': 0.002,
'decay_learning_rate': True,
'nepochs': 1000,
'weight_decay': 0.0,
'clip_thresh': 1.0,
# Save
'checkpoint_interval': 5000,
# Eval:
'max_iters': 200,
'griffin_lim_iters': 60,
'power': 1.5, # Power to raise magnitudes to prior to Griffin-Lim
}
Convert the dictionary to a namespace
hparams = types.SimpleNamespace(**hparams_dict)
def hparams_debug_string():
hp = [' %s: %s' % (name, hparams[name]) for name in sorted(hparams)]
return 'Hyperparameters:\n' + '\n'.join(hp)
3. go to /content/tacotron_pytorch/lib/tacotron/util/audio.py and change ```np.complex``` in line 70 to ```complex```
4. go to /content/pytorch-dc-tts/datasets/emovdb.py line 45 and change ```np.long``` to ```np.int64```
5. go to /content/pytorch-dc-tts/audio.py line 61 and change it to
est = librosa.stft(X_t, n_fft=hp.n_fft, hop_length=hp.hop_length, win_length=hp.win_length)
6. remove ```%tensorflow_version 1.x``` from the second cell in the colab
now it works although Amused emotion is not sounding correct but i'll update this when i fix it
to run the colab code successfully:
Default hyperparameters:
hparams_dict = {
Comma-separated list of cleaners to run on text prior to training and eval. For non-English
}
Convert the dictionary to a namespace
hparams = types.SimpleNamespace(**hparams_dict)
def hparams_debug_string(): hp = [' %s: %s' % (name, hparams[name]) for name in sorted(hparams)] return 'Hyperparameters:\n' + '\n'.join(hp)
return librosa.istft(spectrogram, hop_length=hp.hop_length, win_length=hp.win_length, window="hann")
est = librosa.stft(X_t, n_fft=hp.n_fft, hop_length=hp.hop_length, win_length=hp.win_length)