Update Benchmark Table

README.md

@@ -41,15 +41,15 @@ Please install:
 
 We evaluated the model against different Arabic-STT Wav2Vec models.
 
-|    | model                                 | using_transliation   |      WER |
-|---:|:--------------------------------------|:---------------------|---------:|
-|  0 | bakrianoo/sinai-voice-ar-stt          | True                 | 0.238001 |
-|  1 | elgeish/wav2vec2-large-xlsr-53-arabic | True                 | 0.266527 |
-|  2 | othrif/wav2vec2-large-xlsr-arabic     | True                 | 0.298122 |
-|  3 | bakrianoo/sinai-voice-ar-stt          | False                | 0.448987 |
-|  4 | othrif/wav2vec2-large-xlsr-arabic     | False                | 0.464004 |
-|  5 | anas/wav2vec2-large-xlsr-arabic       | True                 | 0.506191 |
-|  6 | anas/wav2vec2-large-xlsr-arabic       | False                | 0.622288 |
+|    | Model                                 | [using transliteration](https://pypi.org/project/lang-trans/)   |      WER |      Train Datasets |
+|---:|:--------------------------------------|:---------------------|---------:|---------:|
+|  1 | bakrianoo/sinai-voice-ar-stt          | True                 | 0.238001 |Common Voice 6|
+|  2 | elgeish/wav2vec2-large-xlsr-53-arabic | True                 | 0.266527 |Common Voice 6 + Arabic Speech Corpus|
+|  3 | othrif/wav2vec2-large-xlsr-arabic     | True                 | 0.298122 |Common Voice 6|
+|  4 | bakrianoo/sinai-voice-ar-stt          | False                | 0.448987 |Common Voice 6|
+|  5 | othrif/wav2vec2-large-xlsr-arabic     | False                | 0.464004 |Common Voice 6|
+|  6 | anas/wav2vec2-large-xlsr-arabic       | True                 | 0.506191 |Common Voice 4|
+|  7 | anas/wav2vec2-large-xlsr-arabic       | False                | 0.622288 |Common Voice 4|
 
 
 <details>
@@ -80,8 +80,8 @@ resamplers = {  # all three sampling rates exist in test split
 transformation = jiwer.Compose([
     # normalize some diacritics, remove punctuation, and replace Persian letters with Arabic ones
     jiwer.SubstituteRegexes({
-        r'[auiFNKo\\~_،؟»\\?;:\\-,\\.؛«!"]': "", "\\u06D6": "",
-        r"[\\|\\{]": "A", "p": "h", "ک": "k", "ی": "y"}),
+        r'[auiFNKo\\\\~_،؟»\\\\?;:\\\\-,\\\\.؛«!"]': "", "\\\\u06D6": "",
+        r"[\\\\|\\\\{]": "A", "p": "h", "ک": "k", "ی": "y"}),
     # default transformation below
     jiwer.RemoveMultipleSpaces(),
     jiwer.Strip(),
@@ -274,8 +274,8 @@ test_split = test_split.map(predict, batched=True, batch_size=16, remove_columns
 transformation = jiwer.Compose([
     # normalize some diacritics, remove punctuation, and replace Persian letters with Arabic ones
     jiwer.SubstituteRegexes({
-        r'[auiFNKo\\\\~_،؟»\\\\?;:\\\\-,\\\\.؛«!"]': "", "\\\\u06D6": "",
-        r"[\\\\|\\\\{]": "A", "p": "h", "ک": "k", "ی": "y"}),
+        r'[auiFNKo\\\\\\\\~_،؟»\\\\\\\\?;:\\\\\\\\-,\\\\\\\\.؛«!"]': "", "\\\\\\\\u06D6": "",
+        r"[\\\\\\\\|\\\\\\\\{]": "A", "p": "h", "ک": "k", "ی": "y"}),
     # default transformation below
     jiwer.RemoveMultipleSpaces(),
     jiwer.Strip(),