diff --git a/.gitignore b/.gitignore
index 9fe17bcebb32d017968834acfabd03bbae6a24d7..9e0ff35c82ab66e767b5019d7fb4bad78081bf6e 100644
--- a/.gitignore
+++ b/.gitignore
@@ -3,6 +3,9 @@ __pycache__/
 *.py[cod]
 *$py.class
 
+# vscode
+.vscode
+
 # C extensions
 *.so
 
diff --git a/parakeet/datasets/vctk.py b/parakeet/datasets/vctk.py
new file mode 100644
index 0000000000000000000000000000000000000000..8faad4f38a67f7d8c13a4fb88d007cbaaaab4e82
--- /dev/null
+++ b/parakeet/datasets/vctk.py
@@ -0,0 +1,80 @@
+from pathlib import Path
+import pandas as pd
+from ruamel.yaml import YAML
+import io
+
+import librosa
+import numpy as np
+
+from parakeet.g2p.en import text_to_sequence
+from parakeet.data.dataset import Dataset
+from parakeet.data.datacargo import DataCargo
+from parakeet.data.batch import TextIDBatcher, WavBatcher
+
+class VCTK(Dataset):
+    def __init__(self, root: Path):
+        self.root = root
+        self.text_root = self.root.joinpath("txt")
+        self.wav_root = self.root.joinpath("wav48")
+
+        if not (self.root.joinpath("metadata.csv").exists() and 
+                self.root.joinpath("speaker_indices.yaml").exists()):
+            self._prepare_metadata()
+        self.speaker_indices, self.metadata = self._load_metadata()
+
+    def _load_metadata(self):
+        yaml=YAML(typ='safe')
+        speaker_indices = yaml.load(self.root.joinpath("speaker_indices.yaml"))
+        metadata = pd.read_csv(self.root.joinpath("metadata.csv"), 
+                               sep="|", quoting=3, header=1)
+        return speaker_indices, metadata
+
+    def _prepare_metadata(self):
+        metadata = []
+        speaker_to_index = {}
+        for i, speaker_folder in enumerate(self.text_root.iterdir()):
+            if speaker_folder.is_dir():
+                speaker_to_index[speaker_folder.name] = i
+                for text_file in speaker_folder.iterdir():
+                    if text_file.is_file():
+                        with io.open(str(text_file)) as f:
+                            transcription = f.read().strip()
+                    wav_file = text_file.with_suffix(".wav")
+                    metadata.append((wav_file.name, speaker_folder.name, transcription))
+        metadata = pd.DataFrame.from_records(metadata,
+                                             columns=["wave_file", "speaker", "text"])
+        
+        # save them
+        yaml=YAML(typ='safe')
+        yaml.dump(speaker_to_index, self.root.joinpath("speaker_indices.yaml"))
+        metadata.to_csv(self.root.joinpath("metadata.csv"), 
+                        sep="|", quoting=3, index=False)
+
+    def _get_example(self, metadatum):
+        wave_file, speaker, text = metadatum
+        wav_path = self.wav_root.joinpath(speaker, wave_file)
+        wav, sr = librosa.load(str(wav_path), sr=None)
+        phoneme_seq = np.array(text_to_sequence(text))
+        return wav, self.speaker_indices[speaker], phoneme_seq
+
+    def __getitem__(self, index):
+        metadatum = self.metadata.iloc[index]
+        example = self._get_example(metadatum)
+        return example
+
+    def __len__(self):
+        return len(self.metadata)
+
+    def _batch_examples(self, minibatch):
+        wav_batch, speaker_batch, phoneme_batch = [], [], []
+        for example in minibatch:
+            wav, speaker_id, phoneme_seq = example
+            wav_batch.append(wav)
+            speaker_batch.append(speaker_id)
+            phoneme_batch.append(phoneme_seq)
+        wav_batch = WavBatcher(pad_value=0.)(wav_batch)
+        speaker_batch = np.array(speaker_batch)
+        phoneme_batch = TextIDBatcher(pad_id=0)(phoneme_batch)
+        return wav_batch, speaker_batch, phoneme_batch
+
+    
\ No newline at end of file
diff --git a/setup.py b/setup.py
index c6e014800b1d0332cbecb2c3abb253f34f2a218b..2a164465c8b7d8c23c5382458bbd4a80a57c2b0a 100644
--- a/setup.py
+++ b/setup.py
@@ -34,7 +34,9 @@ setup_info = dict(
     license='Apache 2',
 
     install_requires=[
-        'numpy', 'nltk', 'inflect', 'librosa', 'unidecode', 'numba', 'tqdm', 'matplotlib', 'tensorboardX', 'tensorboard', 'scipy', 'tqdm',
+        'numpy', 'nltk', 'inflect', 'librosa', 'unidecode', 'numba', 
+        'tqdm', 'matplotlib', 'tensorboardX', 'tensorboard', 'scipy',
+        'ruamel.yaml', 'pandas',
     ],
 
     # Package info
diff --git a/tests/test_ljspeech.py b/tests/test_ljspeech.py
index 5f837221bdb921a1cb49c31474ba619d7f10ed24..04db6a9997df0ce2c64a472a75a16cdabfd58df8 100644
--- a/tests/test_ljspeech.py
+++ b/tests/test_ljspeech.py
@@ -3,7 +3,7 @@ from parakeet.data.datacargo import DataCargo
 
 from pathlib import Path
 
-LJSPEECH_ROOT = Path("/Users/chenfeiyu/projects/LJSpeech-1.1")
+LJSPEECH_ROOT = Path("/workspace/datasets/LJSpeech-1.1")
 ljspeech = LJSpeech(LJSPEECH_ROOT)
 ljspeech_cargo = DataCargo(ljspeech, batch_size=16, shuffle=True)
 for i, batch in enumerate(ljspeech_cargo):
diff --git a/tests/test_vctk.py b/tests/test_vctk.py
new file mode 100644
index 0000000000000000000000000000000000000000..3f7d61e3c67954e0cb746f60cefa1211e20998a3
--- /dev/null
+++ b/tests/test_vctk.py
@@ -0,0 +1,11 @@
+from parakeet.datasets import vctk
+from pathlib import Path
+from parakeet.data.datacargo import DataCargo
+
+root = Path("/workspace/datasets/VCTK-Corpus")
+vctk_dataset = vctk.VCTK(root)
+vctk_cargo = DataCargo(vctk_dataset, batch_size=16, shuffle=True, drop_last=True)
+
+for i, batch in enumerate(vctk_cargo):
+    print(i)
+