diff --git a/README.md b/README.md
index ec81120..659cb4c 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,20 @@
-# IndicOOV
\ No newline at end of file
+# IndicOOV
+
+## Abstract
+Publicly available TTS datasets for low-resource languages like
+Hindi and Tamil typically contain 10-20 hours of data, lead-
+ing to poor vocabulary coverage. This limitation becomes evi-
+dent in downstream applications where domain-specific vocab-
+ulary coupled with frequent code-mixing with English, results
+in many OOV words. To highlight this problem, we create a
+benchmark containing OOV words from several real-world ap-
+plications. Indeed, state-of-the-art Hindi and Tamil TTS sys-
+tems perform poorly on this OOV benchmark, as indicated by
+intelligibility tests. To improve the model’s OOV performance,
+we propose a low-effort and economically viable strategy to ob-
+tain more training data. Specifically, we propose using volun-
+teers as opposed to high quality voice artists to record words
+containing character bigrams unseen in the training data. We
+show that using such inexpensive data, the model’s performance
+improves on OOV words, while not affecting voice quality and
+in-domain performance.