tensorflow1系を入れずにmellotronを動かす方法

Texarを以下の手順でインストール

git clone https://github.com/asyml/texar-pytorch.git 
cd texar-pytorch
pip install -e .

公式ドキュメントより引用

hparams.pyを以下のように書き換える。（ファイルのパスは自分で設定をしてください）

from text.symbols import symbols
from texar.torch import HParams
# main.py


def create_hparams(hparams_string=None, verbose=False):
    """Create model hyperparameters. Parse nondefault from given string."""

    hparams = {
        ################################
        # Experiment Parameters        #
        ################################
        "epochs": 50000,
        "iters_per_checkpoint": 500,
        "seed": 1234,
        "dynamic_loss_scaling": True,
        "fp16_run": False,
        "distributed_run": False,
        "dist_backend": "nccl",
        "dist_url": "tcp://localhost:54321",
        "cudnn_enabled": True,
        "cudnn_benchmark": False,
        "ignore_layers": ['speaker_embedding.weight'],

        ################################
        # Data Parameters             #
        ################################
        "training_files": $YOUR_PATH,
        "validation_files": $YOUR_PATH,
        "text_cleaners": ['basic_cleaners'],
        "p_arpabet": 1.0,
        "cmudict_path": "data/cmu_dictionary",

        ################################
        # Audio Parameters             #
        ################################
        "max_wav_value": 32768.0,
        "sampling_rate": 22050,
        "filter_length": 1024,
        "hop_length": 256,
        "win_length": 1024,
        "n_mel_channels": 80,
        "mel_fmin": 0.0,
        "mel_fmax": 8000.0,
        "f0_min": 80,
        "f0_max": 880,
        "harm_thresh": 0.25,

        ################################
        # Model Parameters             #
        ################################
        "n_symbols": len(symbols),
        "symbols_embedding_dim": 512,

        # Encoder parameters
        "encoder_kernel_size": 5,
        "encoder_n_convolutions": 3,
        "encoder_embedding_dim": 512,

        # Decoder parameters
        "n_frames_per_step": 1,  # currently only 1 is supported
        "decoder_rnn_dim": 1024,
        "prenet_dim": 256,
        "prenet_f0_n_layers": 1,
        "prenet_f0_dim": 1,
        "prenet_f0_kernel_size": 1,
        "prenet_rms_dim": 0,
        "prenet_rms_kernel_size": 1,
        "max_decoder_steps": 1000,
        "gate_threshold": 0.5,
        "p_attention_dropout": 0.1,
        "p_decoder_dropout": 0.1,
        "p_teacher_forcing": 1.0,

        # Attention parameters
        "attention_rnn_dim": 1024,
        "attention_dim": 128,

        # Location Layer parameters
        "attention_location_n_filters": 32,
        "attention_location_kernel_size": 31,

        # Mel-post processing network parameters
        "postnet_embedding_dim": 512,
        "postnet_kernel_size": 5,
        "postnet_n_convolutions": 5,

        # Speaker embedding
        "n_speakers": 123,
        "speaker_embedding_dim": 128,

        # Reference encoder
        "with_gst": True,
        "ref_enc_filters": [32, 32, 64, 64, 128, 128],
        "ref_enc_size": [3, 3],
        "ref_enc_strides": [2, 2],
        "ref_enc_pad": [1, 1],
        "ref_enc_gru_size": 128,

        # Style Token Layer
        "token_embedding_size": 256,
        "token_num": 10,
        "num_heads": 8,

        ################################
        # Optimization Hyperparameters #
        ################################
        "use_saved_learning_rate": False,
        "learning_rate": 1e-3,
        "learning_rate_min": 1e-5,
        "learning_rate_anneal": 50000,
        "weight_decay": 1e-6,
        "grad_clip_thresh": 1.0,
        "batch_size": 32,
        "mask_padding": True,  # set model's padded outputs to padded values
    }
    """
    if hparams_string:
        tf.compat.v1.logging.info(
            'Parsing command line hparams: %s', hparams_string)
        hparams.parse(hparams_string)

    if verbose:
        tf.compat.v1.logging.info('Final parsed hparams: %s', hparams.values())
    """
    hparams = HParams(hparams, hparams)
    return hparams