correction on trasncription

evaluate_asr.py

@@ -19,8 +19,8 @@ def evaluate_model(ds, pipe, wer_metric):
     for idx, sample in enumerate(tqdm(ds, desc="Evaluating", total=len(list(ds)))):
         audio_sample = sample["audio"]
         transcription = pipe(audio_sample["array"])['text']
-        # Remove , and . from the transcription
-        transcription = transcription.replace(",", "").replace(".", "")
+        # Keep only letter and spaces for evaluation
+        transcription = "".join([char for char in transcription if char.isalpha() or char.isspace()])
         wer = wer_metric.compute(predictions=[transcription.upper()], references=[sample["text"].upper()])
         wer_scores.append(wer)
         wer_results.append({