diff --git a/doc/data.md b/doc/data.md
index 5f4099ae664790b3ab1f4327e6f7850d0257bbde..5584a54fffd00bc7d0bf0efca9dd5eef787dc2b9 100644
--- a/doc/data.md
+++ b/doc/data.md
@@ -44,13 +44,13 @@ Note that filter is applied to all the examples in the base dataset when initial
 
 ### CacheDataset
 
-By default, we preprocess dataset lazily in `DatasetMixin.get_example`. An example is preprocessed only only requested. But `CacheDataset` caches a base dataset (preprocesses all examples and cache s them eagerly beforehand). When preprocessing the dataset is slow, you can use Cachedataset to speed it up, but caching may consume a lot of RAM if the dataset is large.
+By default, we preprocess dataset lazily in `DatasetMixin.get_example`. An example is preprocessed only only requested. But `CacheDataset` caches the base dataset lazily, so each example is processed only once when it is first requested. When preprocessing the dataset is slow, you can use Cachedataset to speed it up, but caching may consume a lot of RAM if the dataset is large.
 
-Finally, if preprocessing the dataset is slow and the processed dataset is too large to cache, you can write your own code to save them into files or databases and defines a Dataset to load them. Dataset is flexible, so you can create your own dataset painlessly.
+Finally, if preprocessing the dataset is slow and the processed dataset is too large to cache, you can write your own code to save them into files or databases, and then define a Dataset to load them. `Dataset` is flexible, so you can create your own dataset painlessly.
 
 ## DataCargo
 
-`DataCargo`, like `Dataset`, is an iterable of batches. We need datacargo because in deep learning, batching examples into batches exploits the computational resources of modern hardwares. You can iterate over it by `iter(datacargo)` or `for batch in datacargo`. It is an iterable but not an iterator, in that in can be iterated more than once.
+`DataCargo`, like `Dataset`, is an iterable of batches. We need datacargo because in deep learning, batching examples into batches exploits the computational resources of modern hardwares. You can iterate it by `iter(datacargo)` or `for batch in datacargo`. `DataCargo` is an iterable but not an iterator, in that in can be iterated more than once.
 
 ### batch function
 
@@ -104,7 +104,7 @@ Dataset   -->  Iterable[Example]  | iter(Dataset)    ->  Iterator[Example]
 DataCargo -->  Iterable[Batch]    | iter(DataCargo)  ->  Iterator[Batch]
 ```
 
- In order to construct an iterator of batches from an iterator of examples, we construct a DataCargo from a Dataset.
+In order to construct an iterator of batches from an iterator of examples, we construct a DataCargo from a Dataset.
 
 
 
diff --git a/parakeet/datasets/ljspeech.py b/parakeet/datasets/ljspeech.py
index 62209e9219a50855fbd94f7deab96537407c4d18..3ab8ac92edd06b5411033ea9407545aa6ee6dd75 100644
--- a/parakeet/datasets/ljspeech.py
+++ b/parakeet/datasets/ljspeech.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from pathlib import Path
+import os
 import numpy as np
 import pandas as pd
 import librosa
@@ -27,13 +27,11 @@ from ..data.batch import TextIDBatcher, SpecBatcher
 class LJSpeech(DatasetMixin):
     def __init__(self, root):
         super(LJSpeech, self).__init__()
-        assert isinstance(root, (
-            str, Path)), "root should be a string or Path object"
-        self.root = root if isinstance(root, Path) else Path(root)
+        self.root = root
         self.metadata = self._prepare_metadata()
 
     def _prepare_metadata(self):
-        csv_path = self.root.joinpath("metadata.csv")
+        csv_path = os.path.join(self.root, "metadata.csv")
         metadata = pd.read_csv(
             csv_path,
             sep="|",
@@ -51,7 +49,7 @@ class LJSpeech(DatasetMixin):
         """
 
         fname, raw_text, normalized_text = metadatum
-        wav_path = self.root.joinpath("wavs", fname + ".wav")
+        wav_path = os.path.join(self.root, "wavs", fname + ".wav")
 
         # load -> trim -> preemphasis -> stft -> magnitude -> mel_scale -> logscale -> normalize
         wav, sample_rate = librosa.load(
diff --git a/parakeet/models/waveflow/data.py b/parakeet/models/waveflow/data.py
index 0c1e914271214de93de514098872183b94843375..83438f79d6618e21483fa583f86946fc5cf9e023 100644
--- a/parakeet/models/waveflow/data.py
+++ b/parakeet/models/waveflow/data.py
@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
 import random
 
 import librosa
@@ -32,7 +33,7 @@ class Dataset(ljspeech.LJSpeech):
 
     def _get_example(self, metadatum):
         fname, _, _ = metadatum
-        wav_path = self.root.joinpath("wavs", fname + ".wav")
+        wav_path = os.path.join(self.root, "wavs", fname + ".wav")
 
         loaded_sr, audio = read(wav_path)
         assert loaded_sr == self.config.sample_rate