diff --git a/README.md b/README.md
index 21dde55..fa5a831 100644
--- a/README.md
+++ b/README.md
@@ -1,12 +1,15 @@
 # occm
+My attempt to employ one-class learning to detect synthetic speech.
+In fancy terms, I am trying to transform a hyper-plane classifier into a sphere covering positive samples.
 
-
+For feature extraction (frontend), I use wav2vec model from Meta and finetune it with a subset of real/synthetic samples.
+For classification (backend), I work with several models SE-Resnet, AASIST, etc. 
+This work is under progress. All suggestions are welcome.
 
 # Note
 
-install this version of fairseq
+## install this version of fairseq
 `pip install git+https://github.com/facebookresearch/fairseq.git@a54021305d6b3c4c5959ac9395135f63202db8f1`
 
-use numpy
-
-float -> float64
\ No newline at end of file
+## fix numpy package if necessary
+float -> float64