add eval scripts

eval.py

@@ -51,7 +51,18 @@ def normalize_text(text: str) -> str:
 
     chars_to_ignore_regex = '[,?.!\-\;\:"“%‘”�—’…–]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
 
-    text = re.sub(chars_to_ignore_regex, "", text.lower())
+    text = re.sub(chars_to_ignore_regex, "", text.lower()) \
+        .replace("\\\\punkt", "") \
+        .replace("\\\\komma", "") \
+        .replace("è", "e") \
+        .replace("é", "e") \
+        .replace("î", "i") \
+        .replace("ü", "u") \
+        .replace("ÿ", "y") \
+        .replace("ô", "o") \
+        .replace("\\", "") \
+        .replace("/", "") \
+        .replace("|", "")
 
     # In addition, we can normalize the target text, e.g. removing new lines characters etc...
     # note that order is important here!

run_eval_cv.sh

@@ -0,0 +1,7 @@
+python eval.py \
+    --model_id marinone94/xls-r-300m-sv-robust \
+    --dataset mozilla-foundation/common_voice_8_0 \
+    --config sv-SE \
+    --split test \
+    --chunk_length_s 10 \
+    --stride_length_s 1

run_eval_real_world.sh

@@ -0,0 +1,7 @@
+python eval.py \
+    --model_id marinone94/xls-r-300m-sv-robust \
+    --dataset speech-recognition-community-v2/dev_data \
+    --config sv \
+    --split validation \
+    --chunk_length_s 10 \
+    --stride_length_s 1