fix

examples/csmsc/voc1/README.md

@@ -3,12 +3,15 @@ This example contains code used to train a [parallel wavegan](http://arxiv.org/a
 ## Dataset
 ### Download and Extract
 Download CSMSC from it's [official website](https://test.data-baker.com/data/index/TNtts/) and extract it to `~/datasets`. Then the dataset is in the directory `~/datasets/BZNSYP`.
-The structure of the folder is listed below.
-```text
-datasets/BZNSYP
-└── Wave
-    └──XXXX.wav files
-```
+datasets/BZNSYP should have three folders:
+
+└─ Wave
+  └─ .wav files (audio speech)
+  └─ PhoneLabeling
+      └─ .interval files (alignment between phoneme and duration)
+  └─ ProsodyLabeling
+    └─ 000001-010000.txt (text with prosodic by pinyin)
+Still we only use .wav files in training.
 
 ### Get MFA Result and Extract
 We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) results to cut silence at the edge of audio.