diff --git a/README.md b/README.md
index fed32599..01ba258d 100644
--- a/README.md
+++ b/README.md
@@ -26,6 +26,7 @@ You can try the real-time end-to-end text-to-speech demonstration in Google Cola
 
 ## What's new
 
+- 2021/10/21 Single-speaker Korean recipe [egs/kss/voc1] is available.
 - 2021/08/24 Add more pretrained models of StyleMelGAN and HiFi-GAN.
 - 2021/08/07 Add initial pretrained models of StyleMelGAN and HiFi-GAN.
 - 2021/08/03 Support [StyleMelGAN](https://arxiv.org/abs/2011.01557) generator and discriminator!
diff --git a/egs/README.md b/egs/README.md
index 428bb53f..88067da5 100644
--- a/egs/README.md
+++ b/egs/README.md
@@ -12,6 +12,7 @@ Currently, the following recipes are supported.
 - [VCTK](https://homepages.inf.ed.ac.uk/jyamagis/page3/page58/page58.html): English multi-speaker
 - [LibriTTS](https://arxiv.org/abs/1904.02882): English multi-speaker
 - [YesNo](https://arxiv.org/abs/1904.02882): English speaker (For debugging)
+- [KSS](https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset): Single Korean female speaker
 
 
 ## How to run the recipe
diff --git a/egs/kss/voc1/cmd.sh b/egs/kss/voc1/cmd.sh
new file mode 100755
index 00000000..19f34210
--- /dev/null
+++ b/egs/kss/voc1/cmd.sh
@@ -0,0 +1,91 @@
+# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
+# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
+# e.g.
+#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
+#
+# Options:
+#   --time <time>: Limit the maximum time to execute.
+#   --mem <mem>: Limit the maximum memory usage.
+#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
+#   --num-threads <ngpu>: Specify the number of CPU core.
+#   --gpu <ngpu>: Specify the number of GPU devices.
+#   --config: Change the configuration file from default.
+#
+# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
+# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
+# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
+# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
+#
+# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
+# These options are mapping to specific options for each backend and
+# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
+# If jobs failed, your configuration might be wrong for your environment.
+#
+#
+# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
+#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
+# =========================================================~
+
+
+# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
+cmd_backend="local"
+
+# Local machine, without any Job scheduling system
+if [ "${cmd_backend}" = local ]; then
+
+    # The other usage
+    export train_cmd="utils/run.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="utils/run.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="utils/run.pl"
+
+# Local machine, without any Job scheduling system
+elif [ "${cmd_backend}" = stdout ]; then
+
+    # The other usage
+    export train_cmd="utils/stdout.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="utils/stdout.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="utils/stdout.pl"
+
+# "qsub" (SGE, Torque, PBS, etc.)
+elif [ "${cmd_backend}" = sge ]; then
+    # The default setting is written in conf/queue.conf.
+    # You must change "-q g.q" for the "queue" for your environment.
+    # To know the "queue" names, type "qhost -q"
+    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
+
+    export train_cmd="utils/queue.pl"
+    export cuda_cmd="utils/queue.pl"
+    export decode_cmd="utils/queue.pl"
+
+# "sbatch" (Slurm)
+elif [ "${cmd_backend}" = slurm ]; then
+    # The default setting is written in conf/slurm.conf.
+    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
+    # To know the "partion" names, type "sinfo".
+    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
+    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
+
+    export train_cmd="utils/slurm.pl"
+    export cuda_cmd="utils/slurm.pl"
+    export decode_cmd="utils/slurm.pl"
+
+elif [ "${cmd_backend}" = ssh ]; then
+    # You have to create ".queue/machines" to specify the host to execute jobs.
+    # e.g. .queue/machines
+    #   host1
+    #   host2
+    #   host3
+    # Assuming you can login them without any password, i.e. You have to set ssh keys.
+
+    export train_cmd="utils/ssh.pl"
+    export cuda_cmd="utils/ssh.pl"
+    export decode_cmd="utils/ssh.pl"
+
+else
+    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
+    return 1
+fi
diff --git a/egs/kss/voc1/conf/melgan.v3.yaml b/egs/kss/voc1/conf/melgan.v3.yaml
new file mode 100755
index 00000000..94053f9c
--- /dev/null
+++ b/egs/kss/voc1/conf/melgan.v3.yaml
@@ -0,0 +1,129 @@
+# This is the hyperparameter configuration file for MelGAN.
+# Please make sure this is adjusted for the LJSpeech dataset. If you want to
+# apply to the other dataset, you might need to carefully change some parameters.
+# Both generator and discriminator are based on MelGAN but we also use
+# STFT-based auxiliary loss with fixed lr. This configuration requires ~4 GB
+# GPU memory and takes ~4 days on TITAN V.
+# The discriminator loss is not stable as v1, i.e., gradually decreasing both
+# real and fake losses (Also, feature matching loss keeps increasing). But in
+# terms of naturalness, this model is better than v1.
+# If you get unstable results, please increase train_max_steps or use v3.long.
+
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+sampling_rate: 44100     # Sampling rate.
+fft_size: 2048           # FFT size.
+hop_size: 512            # Hop size.
+win_length: null         # Window length.
+                         # If set to null, it will be the same as fft_size.
+window: "hann"           # Window function.
+num_mels: 120             # Number of mel basis.
+fmin: 80                 # Minimum freq in mel basis calculation.
+fmax: 22050               # Maximum frequency in mel basis calculation.
+global_gain_scale: 1.0   # Will be multiplied to all of waveform.
+trim_silence: true       # Whether to trim the start and end of silence.
+trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
+trim_frame_size: 2048    # Frame size in trimming.
+trim_hop_size: 512       # Hop size in trimming.
+format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
+
+###########################################################
+#         GENERATOR NETWORK ARCHITECTURE SETTING          #
+###########################################################
+generator_type: "MelGANGenerator" # Generator type.
+generator_params:
+    in_channels: 120               # Number of input channels.
+    out_channels: 1               # Number of output channels.
+    kernel_size: 7                # Kernel size of initial and final conv layers.
+    channels: 512                 # Initial number of channels for conv layers.
+    upsample_scales: [8, 8, 4, 2] # List of Upsampling scales.
+    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
+    stacks: 3                     # Number of stacks in a single residual stack module.
+    use_weight_norm: True         # Whether to use weight normalization.
+    use_causal_conv: False        # Whether to use causal convolution.
+
+###########################################################
+#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
+###########################################################
+discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
+discriminator_params:
+    in_channels: 1                    # Number of input channels.
+    out_channels: 1                   # Number of output channels.
+    scales: 3                         # Number of multi-scales.
+    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
+    downsample_pooling_params:        # Parameters of the above pooling function.
+        kernel_size: 4
+        stride: 2
+        padding: 1
+        count_include_pad: False
+    kernel_sizes: [5, 3]              # List of kernel size.
+    channels: 16                      # Number of channels of the initial conv layer.
+    max_downsample_channels: 1024     # Maximum number of channels of downsampling layers.
+    downsample_scales: [8, 4, 4, 4]   # List of downsampling scales.
+    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
+    nonlinear_activation_params:      # Parameters of nonlinear activation function.
+        negative_slope: 0.2
+    use_weight_norm: True             # Whether to use weight norm.
+
+###########################################################
+#                   STFT LOSS SETTING                     #
+###########################################################
+stft_loss_params:
+    fft_sizes: [2048, 4096, 1024]  # List of FFT size for STFT-based loss.
+    hop_sizes: [240, 480, 100]     # List of hop size for STFT-based loss
+    win_lengths: [1200, 2400, 480] # List of window length for STFT-based loss.
+    window: "hann_window"         # Window function for STFT-based loss
+
+###########################################################
+#               ADVERSARIAL LOSS SETTING                  #
+###########################################################
+use_feat_match_loss: true # Whether to use feature matching loss.
+lambda_feat_match: 25.0   # Loss balancing coefficient for feature matching loss.
+lambda_adv: 4.0           # Loss balancing coefficient for adversarial loss.
+
+###########################################################
+#                  DATA LOADER SETTING                    #
+###########################################################
+batch_size: 16             # Batch size.
+batch_max_steps: 8192      # Length of each audio in batch. Make sure dividable by hop_size.
+pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
+num_workers: 2             # Number of workers in Pytorch DataLoader.
+remove_short_samples: true # Whether to remove samples the length of which are less than batch_max_steps.
+allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
+
+###########################################################
+#             OPTIMIZER & SCHEDULER SETTING               #
+###########################################################
+generator_optimizer_params:
+    lr: 0.0001             # Generator's learning rate.
+    eps: 1.0e-6            # Generator's epsilon.
+    weight_decay: 0.0      # Generator's weight decay coefficient.
+generator_scheduler_params:
+    step_size: 2000000     # Generator's scheduler step size.
+    gamma: 0.5             # Generator's scheduler gamma.
+                           # At each step size, lr will be multiplied by this parameter.
+generator_grad_norm: 10    # Generator's gradient norm.
+discriminator_optimizer_params:
+    lr: 0.00005            # Discriminator's learning rate.
+    eps: 1.0e-6            # Discriminator's epsilon.
+    weight_decay: 0.0      # Discriminator's weight decay coefficient.
+discriminator_scheduler_params:
+    step_size: 2000000     # Discriminator's scheduler step size.
+    gamma: 0.5             # Discriminator's scheduler gamma.
+                           # At each step size, lr will be multiplied by this parameter.
+discriminator_grad_norm: 1 # Discriminator's gradient norm.
+
+###########################################################
+#                    INTERVAL SETTING                     #
+###########################################################
+discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
+train_max_steps: 1000000                # Number of training steps.
+save_interval_steps: 50000               # Interval steps to save checkpoint.
+eval_interval_steps: 1000               # Interval steps to evaluate the network.
+log_interval_steps: 100                 # Interval steps to record the training log.
+
+###########################################################
+#                     OTHER SETTING                       #
+###########################################################
+num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.
diff --git a/egs/kss/voc1/conf/multi_band_melgan.v1.yaml b/egs/kss/voc1/conf/multi_band_melgan.v1.yaml
new file mode 100755
index 00000000..438341a1
--- /dev/null
+++ b/egs/kss/voc1/conf/multi_band_melgan.v1.yaml
@@ -0,0 +1,147 @@
+# This is the hyperparameter configuration file for MelGAN.
+# Please make sure this is adjusted for the LJSpeech dataset. If you want to
+# apply to the other dataset, you might need to carefully change some parameters.
+# This configuration requires ~ 8GB memory and will finish within 5 days on Titan V.
+
+# This configuration is based on multi-band MelGAN. Hop size and sampling rate are
+# different from the paper (16kHz vs 22.05kHz). And we use RAdam instaed of Adam
+# with batch size = 64 instaed of 128. Tne number of iteraions is now shown in
+# the paper so currently we train 1M iterations (not sure enough to converge).
+
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+sampling_rate: 44100     # Sampling rate.
+fft_size: 2048           # FFT size.
+hop_size: 512            # Hop size.
+win_length: null         # Window length.
+                         # If set to null, it will be the same as fft_size.
+window: "hann"           # Window function.
+num_mels: 120             # Number of mel basis.
+fmin: 80                 # Minimum freq in mel basis calculation.
+fmax: 22050               # Maximum frequency in mel basis calculation.
+global_gain_scale: 1.0   # Will be multiplied to all of waveform.
+trim_silence: true       # Whether to trim the start and end of silence.
+trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
+trim_frame_size: 2048    # Frame size in trimming.
+trim_hop_size: 512       # Hop size in trimming.
+format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
+
+###########################################################
+#         GENERATOR NETWORK ARCHITECTURE SETTING          #
+###########################################################
+generator_type: "MelGANGenerator" # Generator type.
+generator_params:
+    in_channels: 120               # Number of input channels.
+    out_channels: 4               # Number of output channels.
+    kernel_size: 7                # Kernel size of initial and final conv layers.
+    channels: 384                 # Initial number of channels for conv layers.
+    upsample_scales: [8, 4, 4]    # List of Upsampling scales.
+    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
+    stacks: 4                     # Number of stacks in a single residual stack module.
+    use_weight_norm: True         # Whether to use weight normalization.
+    use_causal_conv: False        # Whether to use causal convolution.
+
+###########################################################
+#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
+###########################################################
+discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
+discriminator_params:
+    in_channels: 1                    # Number of input channels.
+    out_channels: 1                   # Number of output channels.
+    scales: 3                         # Number of multi-scales.
+    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
+    downsample_pooling_params:        # Parameters of the above pooling function.
+        kernel_size: 4
+        stride: 2
+        padding: 1
+        count_include_pad: False
+    kernel_sizes: [5, 3]              # List of kernel size.
+    channels: 16                      # Number of channels of the initial conv layer.
+    max_downsample_channels: 512      # Maximum number of channels of downsampling layers.
+    downsample_scales: [8, 4, 4]      # List of downsampling scales.
+    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
+    nonlinear_activation_params:      # Parameters of nonlinear activation function.
+        negative_slope: 0.2
+    use_weight_norm: True             # Whether to use weight norm.
+
+###########################################################
+#                   STFT LOSS SETTING                     #
+###########################################################
+stft_loss_params:
+    fft_sizes: [2048, 4096, 1024]  # List of FFT size for STFT-based loss.
+    hop_sizes: [240, 480, 100]     # List of hop size for STFT-based loss
+    win_lengths: [1200, 2400, 480] # List of window length for STFT-based loss.
+    window: "hann_window"         # Window function for STFT-based loss
+use_subband_stft_loss: true
+subband_stft_loss_params:
+    fft_sizes: [768, 1366, 342]  # List of FFT size for STFT-based loss.
+    hop_sizes: [60, 120, 20]     # List of hop size for STFT-based loss
+    win_lengths: [300, 600, 120] # List of window length for STFT-based loss.
+    window: "hann_window"       # Window function for STFT-based loss
+
+###########################################################
+#               ADVERSARIAL LOSS SETTING                  #
+###########################################################
+use_feat_match_loss: false # Whether to use feature matching loss.
+lambda_adv: 2.5            # Loss balancing coefficient for adversarial loss.
+
+###########################################################
+#                  DATA LOADER SETTING                    #
+###########################################################
+batch_size: 64             # Batch size.
+batch_max_steps: 16384     # Length of each audio in batch. Make sure dividable by hop_size.
+pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
+num_workers: 4             # Number of workers in Pytorch DataLoader.
+remove_short_samples: true # Whether to remove samples the length of which are less than batch_max_steps.
+allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
+
+###########################################################
+#             OPTIMIZER & SCHEDULER SETTING               #
+###########################################################
+generator_optimizer_type: "RAdam"       # Generator's optimizer type.
+generator_optimizer_params:
+    lr: 0.0001                          # Generator's learning rate.
+    eps: 1.0e-6                         # Generator's epsilon.
+    weight_decay: 0.0                   # Generator's weight decay coefficient.
+generator_grad_norm: 10                 # Generator's gradient norm.
+generator_scheduler_type: "MultiStepLR" # Generator's scheduler type.
+generator_scheduler_params:
+    gamma: 0.5                          # Generator's scheduler gamma.
+    milestones:                         # At each milestone, lr will be multiplied by gamma.
+        - 100000
+        - 200000
+        - 300000
+        - 400000
+        - 500000
+        - 600000
+discriminator_optimizer_type: "RAdam"       # Discriminator's optimizer type.
+discriminator_optimizer_params:
+    lr: 0.0001                              # Discriminator's learning rate.
+    eps: 1.0e-6                             # Discriminator's epsilon.
+    weight_decay: 0.0                       # Discriminator's weight decay coefficient.
+discriminator_grad_norm: 1                  # Discriminator's gradient norm.
+discriminator_scheduler_type: "MultiStepLR" # Discriminator's scheduler type.
+discriminator_scheduler_params:
+    gamma: 0.5                              # Discriminator's scheduler gamma.
+    milestones:                             # At each milestone, lr will be multiplied by gamma.
+        - 100000
+        - 200000
+        - 300000
+        - 400000
+        - 500000
+        - 600000
+
+###########################################################
+#                    INTERVAL SETTING                     #
+###########################################################
+discriminator_train_start_steps: 200000 # Number of steps to start to train discriminator.
+train_max_steps: 1000000                # Number of training steps.
+save_interval_steps: 50000              # Interval steps to save checkpoint.
+eval_interval_steps: 1000               # Interval steps to evaluate the network.
+log_interval_steps: 1000                # Interval steps to record the training log.
+
+###########################################################
+#                     OTHER SETTING                       #
+###########################################################
+num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.
diff --git a/egs/kss/voc1/conf/slurm.conf b/egs/kss/voc1/conf/slurm.conf
new file mode 100755
index 00000000..cefd21f0
--- /dev/null
+++ b/egs/kss/voc1/conf/slurm.conf
@@ -0,0 +1,12 @@
+# Default configuration
+command sbatch --export=PATH  --ntasks-per-node=1
+option time=* --time $0
+option mem=* --mem-per-cpu $0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* --cpus-per-task $0 --ntasks-per-node=1
+option num_threads=1 --cpus-per-task 1  --ntasks-per-node=1 # Do not add anything to qsub_opts
+default gpu=0
+option gpu=0 -p cpu
+option gpu=* -p gpu --gres=gpu:$0
+# note: the --max-jobs-run option is supported as a special case
+# by slurm.pl and you don't have to handle it in the config file.
diff --git a/egs/kss/voc1/local/data_prep.sh b/egs/kss/voc1/local/data_prep.sh
new file mode 100755
index 00000000..00e64f4e
--- /dev/null
+++ b/egs/kss/voc1/local/data_prep.sh
@@ -0,0 +1,80 @@
+#!/usr/bin/env bash
+
+# Copyright 2021 Tomoki Hayashi
+#  Apache 2.0  (http://www.apache.org/licenses/LICENSE-2.0)
+
+# shellcheck disable=SC1091
+. ./path.sh || exit 1;
+
+num_dev=250
+num_eval=250
+train_set="train_nodev"
+dev_set="dev"
+eval_set="eval"
+shuffle=false
+
+# shellcheck disable=SC1091
+. utils/parse_options.sh || exit 1;
+
+db_root=$1
+data_dir=$2
+
+# check arguments
+if [ $# != 2 ]; then
+    echo "Usage: $0 [Options] <db_root> <data_dir>"
+    echo "e.g.: $0 downloads/LJSpeech-1.1 data"
+    echo ""
+    echo "Options:"
+    echo "    --num_dev: number of development uttreances (default=250)."
+    echo "    --num_eval: number of evaluation uttreances (default=250)."
+    echo "    --train_set: name of train set (default=train_nodev)."
+    echo "    --dev_set: name of dev set (default=dev)."
+    echo "    --eval_set: name of eval set (default=eval)."
+    echo "    --shuffle: whether to perform shuffle in making dev / eval set (default=false)."
+    exit 1
+fi
+
+set -euo pipefail
+
+# check directory existence
+[ ! -e "${data_dir}" ] && mkdir -p "${data_dir}"
+
+# set filenames
+mkdir -p "${data_dir}/all"
+scp=${data_dir}/all/wav.scp
+
+# check file existence
+[ -e "${scp}" ] && rm "${scp}"
+
+# make scp, utt2spk, and spk2utt
+find "${db_root}/kss" -name "*.wav" | sort | while read -r filename; do
+    id=kss_$(basename "${filename}" | sed -e "s/\.[^\.]*$//g")
+    # NOTE(kan-bayashi): Some wav files are stereo
+    echo "${id} sox ${filename} -t wav -c 1 - |" >> "${scp}"
+done
+echo "Successfully finished making wav.scp."
+
+# split
+num_all=$(wc -l < "${scp}")
+num_deveval=$((num_dev + num_eval))
+num_train=$((num_all - num_deveval))
+utils/split_data.sh \
+    --num_first "${num_train}" \
+    --num_second "${num_deveval}" \
+    --shuffle "${shuffle}" \
+    "${data_dir}/all" \
+    "${data_dir}/${train_set}" \
+    "${data_dir}/deveval"
+utils/split_data.sh \
+    --num_first "${num_dev}" \
+    --num_second "${num_eval}" \
+    --shuffle "${shuffle}" \
+    "${data_dir}/deveval" \
+    "${data_dir}/${dev_set}" \
+    "${data_dir}/${eval_set}"
+
+# remove tmp directories
+rm -rf "${data_dir}/all"
+rm -rf "${data_dir}/deveval"
+
+echo "Successfully prepared data."
diff --git a/egs/kss/voc1/path.sh b/egs/kss/voc1/path.sh
new file mode 100755
index 00000000..b0ca27c6
--- /dev/null
+++ b/egs/kss/voc1/path.sh
@@ -0,0 +1,33 @@
+# cuda related
+export CUDA_HOME=/usr/local/cuda-10.0
+export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
+
+# path related
+export PRJ_ROOT="${PWD}/../../.."
+if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
+    # shellcheck disable=SC1090
+    . "${PRJ_ROOT}/tools/venv/bin/activate"
+fi
+
+# python related
+export OMP_NUM_THREADS=1
+export PYTHONIOENCODING=UTF-8
+export MPL_BACKEND=Agg
+
+# check installation
+if ! command -v parallel-wavegan-train > /dev/null; then
+    echo "Error: It seems setup is not finished." >&2
+    echo "Error: Please setup your environment by following README.md" >&2
+    return 1
+fi
+if ! command -v jq > /dev/null; then
+    echo "Error: It seems jq is not installed." >&2
+    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
+    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
+    return 1
+fi
+if ! command -v yq > /dev/null; then
+    echo "Error: It seems yq is not installed." >&2
+    echo "Error: Please install via \`pip install yq\`." >&2
+    return 1
+fi
diff --git a/egs/kss/voc1/run.sh b/egs/kss/voc1/run.sh
new file mode 100755
index 00000000..04b87bea
--- /dev/null
+++ b/egs/kss/voc1/run.sh
@@ -0,0 +1,158 @@
+#!/bin/bash
+
+# Copyright 2019 Tomoki Hayashi
+#  MIT License (https://opensource.org/licenses/MIT)
+
+. ./cmd.sh || exit 1;
+. ./path.sh || exit 1;
+
+# basic settings
+stage=-1       # stage to start
+stop_stage=100 # stage to stop
+verbose=1      # verbosity level (lower is less info)
+n_gpus=1       # number of gpus in training
+n_jobs=16      # number of parallel jobs in feature extraction
+
+# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
+conf=conf/melgan.v3.yaml
+
+# directory path setting
+download_dir=downloads/archive # direcotry to save downloaded files
+dumpdir=dump           # directory to dump features
+
+# training related setting
+tag=""     # tag for directory to save model
+resume=""  # checkpoint path to resume training
+           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
+
+# decoding related setting
+checkpoint="" # checkpoint path to be used for decoding
+              # if not provided, the latest one will be used
+              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
+
+# shellcheck disable=SC1091
+. utils/parse_options.sh || exit 1;
+
+train_set="train_nodev" # name of training data directory
+dev_set="dev"           # name of development data direcotry
+eval_set="eval"         # name of evaluation data direcotry
+
+set -euo pipefail
+
+if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
+    echo "Stage 0: Data preparation"
+    local/data_prep.sh \
+        --train_set "${train_set}" \
+        --dev_set "${dev_set}" \
+        --eval_set "${eval_set}" \
+        "${download_dir}" data
+fi
+
+stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
+if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
+    echo "Stage 1: Feature extraction"
+    # extract raw features
+    pids=()
+    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
+    (
+        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
+        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
+        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
+        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
+            parallel-wavegan-preprocess \
+                --config "${conf}" \
+                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
+                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
+                --verbose "${verbose}"
+        echo "Successfully finished feature extraction of ${name} set."
+    ) &
+    pids+=($!)
+    done
+    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
+    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
+    echo "Successfully finished feature extraction."
+
+    # calculate statistics for normalization
+    echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
+    ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
+        parallel-wavegan-compute-statistics \
+            --config "${conf}" \
+            --rootdir "${dumpdir}/${train_set}/raw" \
+            --dumpdir "${dumpdir}/${train_set}" \
+            --verbose "${verbose}"
+    echo "Successfully finished calculation of statistics."
+
+    # normalize and dump them
+    pids=()
+    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
+    (
+        [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
+        echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
+        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
+            parallel-wavegan-normalize \
+                --config "${conf}" \
+                --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
+                --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
+                --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
+                --verbose "${verbose}"
+        echo "Successfully finished normalization of ${name} set."
+    ) &
+    pids+=($!)
+    done
+    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
+    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
+    echo "Successfully finished normalization."
+fi
+
+if [ -z "${tag}" ]; then
+    expdir="exp/${train_set}_kss_$(basename "${conf}" .yaml)"
+else
+    expdir="exp/${train_set}_kss_${tag}"
+fi
+if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
+    echo "Stage 2: Network training"
+    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
+    cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
+    if [ "${n_gpus}" -gt 1 ]; then
+        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
+    else
+        train="parallel-wavegan-train"
+    fi
+    echo "Training start. See the progress via ${expdir}/train.log."
+    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
+        ${train} \
+            --config "${conf}" \
+            --train-dumpdir "${dumpdir}/${train_set}/norm" \
+            --dev-dumpdir "${dumpdir}/${dev_set}/norm" \
+            --outdir "${expdir}" \
+            --resume "${resume}" \
+            --verbose "${verbose}"
+    echo "Successfully finished training."
+fi
+
+if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
+    echo "Stage 3: Network decoding"
+    # shellcheck disable=SC2012
+    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
+    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
+    pids=()
+    for name in "${dev_set}" "${eval_set}"; do
+    (
+        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
+        [ "${n_gpus}" -gt 1 ] && n_gpus=1
+        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
+        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
+            parallel-wavegan-decode \
+                --dumpdir "${dumpdir}/${name}/norm" \
+                --checkpoint "${checkpoint}" \
+                --outdir "${outdir}/${name}" \
+                --verbose "${verbose}"
+        echo "Successfully finished decoding of ${name} set."
+    ) &
+    pids+=($!)
+    done
+    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
+    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
+    echo "Successfully finished decoding."
+fi
+echo "Finished."
diff --git a/egs/kss/voc1/utils b/egs/kss/voc1/utils
new file mode 120000
index 00000000..973afe67
--- /dev/null
+++ b/egs/kss/voc1/utils
@@ -0,0 +1 @@
+../../../utils
\ No newline at end of file