diff --git a/README.md b/README.md
index 7785e035ebdef0a16b70c24038c8e4a912406c96..50c64eea0c772ff775b58f1c78b1120751849b20 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,23 @@
 # Cross-species F0 estimation, dataset and study of baseline algorithms
 
 ## Use a crepe model pretrained on animal signals to analyse your own signals
-TODO
+Use the `predict.py` script to run a pretrained crepe model to estimate fundamental frequency values for your own sounds.
+```
+usage: predict.py [-h] [--model_path MODEL_PATH] [--compress COMPRESS] [--step STEP] [--decoder {argmax,weighted_argmax,viterbi}] [--print PRINT] indir
+
+positional arguments:
+  indir                 Directory with sound files to process
+
+optional arguments:
+  -h, --help            show this help message and exit
+  --model_path MODEL_PATH
+                        Path of model weights
+  --compress COMPRESS   Compression factor used to shift frequencies into CREPE's range [32Hz; 2kHz]. Frequencies are divided by the given factor by artificially changing the sampling rate (slowing down / speeding up the signal).
+  --step STEP           Step used between each prediction (in seconds)
+  --decoder {argmax,weighted_argmax,viterbi}
+                        Decoder used to postprocess predictions
+  --print PRINT         Print spectrograms with overlaid F0 predictions to assess their quality
+```
 
 ## Reproduce paper experiments