boumehdi commited on
Commit
bda656c
1 Parent(s): be3434d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -6
README.md CHANGED
@@ -17,11 +17,11 @@ model-index:
17
  metrics:
18
  - name: Test WER
19
  type: wer
20
- value: 49.68
21
  ---
22
  # Wav2Vec2-Large-XLSR-53-Moroccan-Darija
23
 
24
- **wav2vec2-large-xlsr-53** fine-tuned on 6 hours of labeled Darija Audios
25
 
26
  I have also added 3 phonetic units to this model ڭ, ڤ and پ. For example: ڭال , ڤيديو , پودكاست
27
 
@@ -59,7 +59,17 @@ print(transcription)
59
 
60
  Here's the output: ڭالت ليا هاد السيد هادا ما كاينش بحالو
61
 
62
- ## Evaluation
 
 
 
 
 
 
 
 
 
 
63
 
64
  **Wer**: 49.68
65
 
@@ -67,10 +77,10 @@ Here's the output: ڭالت ليا هاد السيد هادا ما كاينش ب
67
 
68
  **Validation Loss**: 45.24
69
 
70
- This high validation loss value is mainly due to the fact that Darija can be written in many ways.
71
-
72
  ## Future Work
73
 
74
- Currently working on improving this model. The new model will be available soon.
 
 
75
 
76
 
 
17
  metrics:
18
  - name: Test WER
19
  type: wer
20
+ value: 44.30
21
  ---
22
  # Wav2Vec2-Large-XLSR-53-Moroccan-Darija
23
 
24
+ **wav2vec2-large-xlsr-53** fine-tuned on 8.5 hours of labeled Darija Audios
25
 
26
  I have also added 3 phonetic units to this model ڭ, ڤ and پ. For example: ڭال , ڤيديو , پودكاست
27
 
 
59
 
60
  Here's the output: ڭالت ليا هاد السيد هادا ما كاينش بحالو
61
 
62
+ ## Evaluation & Previous works
63
+
64
+ -v2 (fine-tuned on 8.5 hours of audio + replacing أ and ى and إ with ا + tried to standardize the Moroccan Darija)
65
+
66
+ **Wer**: 44.30
67
+
68
+ **Training Loss**: 12.99
69
+
70
+ **Validation Loss**: 36.93
71
+
72
+ -v1 (fine-tuned on 6 hours of audio)
73
 
74
  **Wer**: 49.68
75
 
 
77
 
78
  **Validation Loss**: 45.24
79
 
 
 
80
  ## Future Work
81
 
82
+ I am currently working on improving this model. The new model will be available soon.
83
+
84
+ email: souregh@gmail.com
85
 
86