dafault.yaml

---
default:
    sr: 16000
    frame_shift: 0.005
    frame_length:  0.025
    win_length: 400
    hop_length: 80
    n_fft: 512
    preemphasis: 0.97
    n_mfcc: 40
    n_iter: 60 # Number of inversion iterations
    n_mels: 80
    duration: 2
    max_db: 35
    min_db: -55

    # model
    hidden_units: 256  # alias: E
    num_banks: 16
    num_highway_blocks: 4
    norm_type: 'ins'  # a normalizer function. value: bn, ln, ins, or None
    t: 1.0  # temperature
    dropout_rate: 0.2

    # train
    batch_size: 32

logdir_path: '/data/private/vc/logdir'

train1:
    # path
    data_path: '/content/deep-voice-conversion/timit/data/TRAIN/*/*/*.wav'

    # model
    hidden_units: 128  # alias: E
    num_banks: 8
    num_highway_blocks: 4
    norm_type: 'ins'  # a normalizer function. value: bn, ln, ins, or None
    t: 1.0  # temperature
    dropout_rate: 0.2

    # train
    batch_size: 32
    lr: 0.0003
    num_epochs: 1000
    steps_per_epoch: 100
    save_per_epoch: 2
    num_gpu: 2
---
train2:
    # path
    data_path: '/data/private/vc/datasets/arctic/slt/*.wav'

    # model
    hidden_units: 256  # alias: E
    num_banks: 8
    num_highway_blocks: 8
    norm_type: 'ins'  # a normalizer function. value: bn, ln, ins, or None
    t: 1.0  # temperature
    dropout_rate: 0.2

    # train
    batch_size: 32
    lr: 0.0003
    lr_cyclic_margin: 0.
    lr_cyclic_steps: 5000
    clip_value_max: 3.
    clip_value_min: -3.
    clip_norm: 10
    num_epochs: 10000
    steps_per_epoch: 100
    save_per_epoch: 50
    test_per_epoch: 1
    num_gpu: 4
---
test1:
    # path
    data_path: '/data/private/vc/datasets/timit/TIMIT/TEST/*/*/*.wav'

    # test
    batch_size: 32
---
test2:
    # path
    data_path: '/data/private/vc/datasets/arctic/slt/*.wav'

    # test
    batch_size: 32
---
convert:
    # path
    data_path: '/data/private/vc/datasets/arctic/bdl/*.wav'

    # convert
    one_full_wav: False
    batch_size: 1
    emphasis_magnitude: 1.2

gistfile1.txt

! python train1.py  -gpu 0 test124

OJ_deep_voice_conversion.ipynb

train1.py

# -*- coding: utf-8 -*-
# /usr/bin/python2
import argparse
import multiprocessing
import os

from tensorpack.callbacks.saver import ModelSaver
from tensorpack.tfutils.sessinit import SaverRestore
from tensorpack.train.interface import TrainConfig
from tensorpack.train.interface import launch_train_with_config
from tensorpack.train.trainers import *
from tensorpack.utils import logger
from tensorpack.input_source.input_source import QueueInput
from data_load import Net1DataFlow
from hparam import hparam as hp
from models import Net1
import tensorflow as tf


def train(args, logdir):

    # model
    model = Net1()

    # dataflow
    df = Net1DataFlow(hp.train1.data_path, hp.train1.batch_size)

    # set logger for event and model saver
    logger.set_logger_dir(logdir)

    session_conf = tf.ConfigProto(
        gpu_options=tf.GPUOptions(
            allow_growth=True,
        ),)

    train_conf = TrainConfig(
        model=model,
        data=QueueInput(df(n_prefetch=1000, n_thread=4)),
        callbacks=[
            ModelSaver(checkpoint_dir=logdir),
            # TODO EvalCallback()
        ],
        max_epoch=hp.train1.num_epochs,
        steps_per_epoch=hp.train1.steps_per_epoch,
        # session_config=session_conf
    )
    ckpt = '{}/{}'.format(logdir, args.ckpt) if args.ckpt else tf.train.latest_checkpoint(logdir)
    if ckpt:
        train_conf.session_init = SaverRestore(ckpt)

    if args.gpu:
        os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
        #train_conf.nr_tower = len(args.gpu.split(','))

    trainer = SimpleTrainer()

    launch_train_with_config(train_conf, trainer=trainer)


def get_arguments():
    parser = argparse.ArgumentParser()
    parser.add_argument('case', type=str, help='experiment case name')
    parser.add_argument('-ckpt', help='checkpoint to load model.')
    parser.add_argument('-gpu', help='comma separated list of GPU(s) to use.')
    arguments = parser.parse_args()
    return arguments

if __name__ == '__main__':
    args = get_arguments()
    hp.set_hparam_yaml(args.case)
    logdir_train1 = '{}/train1'.format(hp.logdir)

    print('case: {}, logdir: {}'.format(args.case, args.case, logdir_train1))

    train(args, logdir=logdir_train1)

    print("Done")