making doc clearer

facebookresearch · Sep 7, 2023 · 5e935bc · 5e935bc
1 parent 3c8c1d0
commit 5e935bc
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 3 deletions.
diff --git a/audiocraft/losses/sisnr.py b/audiocraft/losses/sisnr.py
@@ -41,8 +41,9 @@ class SISNR(nn.Module):
 
     Input should be [B, C, T], output is scalar.
 
-    ..Warning:: We report here minus SISNR, so negative values are better. This is so that
-        this can be naturally used as a loss!
+    ..Warning:: This function returns the opposite of the SI-SNR (e.g. `-1 * regular_SI_SNR`). Consequently, lower scores are
+        better in terms of reconstruction quality, in particular, it should be negative if training
+        goes well. This done this way so that this module can also be used as a loss function for training model.
 
     Args:
         sample_rate (int): Sample rate.

diff --git a/docs/METRICS.md b/docs/METRICS.md
@@ -16,7 +16,9 @@ We provide an implementation of the Scale-Invariant Signal-to-Noise Ratio in PyT
 No specific requirement is needed for this metric. Please activate the metric at the
 evaluation stage with the appropriate flag:
 
-**Warning:** As we also allow SISNR to be used as a training loss, we actually report minus the SISNR, e.g. negative values are best.
+**Warning:** We report the opposite of the SI-SNR, e.g. multiplied by -1. This is due to internal 
+    details where the SI-SNR score can also be used as a training loss function, where lower
+    values should indicate better reconstruction. Negative values are such expected and a good sign! Those should be again multiplied by `-1` before publication :)
 
 ```shell
 dora run <...> evaluate.metrics.sisnr=true