Update README.md

README.md

@@ -17,13 +17,11 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 22.56
+         value: 0.09
 ---
 # Wav2Vec2-Large-XLSR-53-Moroccan-Darija
 
-**wav2vec2-large-xlsr-53** fine-tuned on 11 hours (~79 speakers) of labeled Darija Audios
-
-The vocabulary contains 3 additional phonetic units ڭ, ڤ and پ. For example: ڭال , ڤيديو , پودكاست
+**wav2vec2-large-xlsr-53** fine-tuned on 120 hours of labeled Darija Audios
 
 ## Usage
 
@@ -63,7 +61,15 @@ Here's the output: ڭالت ليا هاد السيد هادا ما كاينش ب
 
 ====================================
 
--v4 (working on it, it's going to take some time)
+-v5 working on it and might add some common french words (mais, alors, donc, ...)
+
+-v4 
+
+**Wer**: 0.09
+
+**Training Loss**: 13.59
+
+**Validation Loss**: 0.07
 
 -v3 (fine-tuned on 11 hours of audio + changed hyperparameters + discovered a huge mistake when using the letter 'ا' that has improved the WER dramatically)