diff --git a/README.md b/README.md
index a0a6a49..32bceef 100644
--- a/README.md
+++ b/README.md
@@ -116,7 +116,7 @@ Logging arguments:
 
 After training, you can evaluate the model in terms of lingustic content (WER and CER) and target characteristic (SV).
 
-You need to keep the model arguments in the training phase. the code only supports the version in which the number of target utterances is 1.
+You need to keep the model arguments in the training phase. The code only supports the version in which the number of target utterances is 1.
 ```
 python main.py test
 evaluation arguments:
@@ -140,25 +140,9 @@ We provide two versions of models depending on input types (mel, cpc).
 
 ## 4. Custom convert
 
-For custom enhancement, you can estimate enhanced speech by running code with ```custom_enhance.py```. 
-The codes include input data processing (downsample from 48 kHz to 16 kHz).
-```
-python custom_enhance.py
-
-enhance arguments:
-  --device: Cuda device or CPU
-  --noisy_path: Path (folder) which contains noisy wav files
-  --model_name: Model version (small, base, large) - you can use one of them
-```
-
-## 5. Converted samples
-
-You can find converted samples in ```./samples``` or please visit our [demo site](https://winddori2002.github.io/vc-demo.github.io/).
-
-In ```./samples```, they are generated by "TriAAN-VC-CPC". 
-
-For sampels in demo site, they are generated by paper version ("TriAAN-VC-CPC")
-
+For custom conversion, you can run the code with ```convert.py```. 
+The codes include data processing, predicting, and vocoding.
+You can find converted examples in ```./samples``` or please visit our [demo site](https://winddori2002.github.io/vc-demo.github.io/).
 
 # Experimental Results