variants

README.md

@@ -62,15 +62,16 @@ the [VoxCeleb1-clean test set](https://www.robots.ox.ac.uk/~vgg/data/voxceleb/me
 This value can be interpreted as the ability to identify speakers only with timbral cues. 
 Tests between two utterances leading to a cosine similarity above the threshold should be considered as similar in terms of timbral cues.
 
+A discussion about this interpretation can be
+found in the paper mentioned hereafter, as well as other experiments showing correlations between these embeddings and timbral voice attributes.
+
 The table below provides the EER and threshold of the different [variants](#variants) of this model.
 
 | Variant name| EER (%) | threshold |
 | --- | --- | --- |
 | W-TBR   | 1.68 | 0.472 |
 | WTA128 | 2.29 | 0.472 |
-
-A discussion about this interpretation can be
-found in the paper mentioned hereafter, as well as other experiments showing correlations between these embeddings and timbral voice attributes.
+| WTA64 | 2.88 | 0.446 |
 
 Please note that the EER value can vary a little depending on the max_size defined to reduce long audios (max 30 seconds in our case).
 
@@ -114,6 +115,7 @@ The table below provides a short description of the variants and their performan
 | --- | --- | --- | --- | 
 | W-TBR | main    | baseline, description in paper | 128 | 
 | WTA128 | wta128 | enriched training dataset, more conversions | 128  | 
+| WTA64 | wta64 | enriched training dataset, more conversions | 64  | 
 
 
 # License