Update README.md

README.md

@@ -76,7 +76,7 @@ from jiwer import wer
 librispeech_eval = load_dataset("librispeech_asr", "clean", split="test")
 
 model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h-lv60-self").to("cuda")
-tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-base-960h-lv60-self")
+tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
 
 def map_to_array(batch):
     speech, _ = sf.read(batch["file"])
@@ -86,9 +86,12 @@ def map_to_array(batch):
 librispeech_eval = librispeech_eval.map(map_to_array)
 
 def map_to_pred(batch):
-    input_values = tokenizer(batch["speech"], return_tensors="pt", padding="longest").input_values
+    inputs = tokenizer(batch["speech"], return_tensors="pt", padding="longest")
+    input_values = inputs.input_values.to("cuda")
+    attention_mask = inputs.attention_mask.to("cuda")
+    
     with torch.no_grad():
-        logits = model(input_values.to("cuda")).logits
+        logits = model(input_values, attention_mask=attention_mask).logits
 
     predicted_ids = torch.argmax(logits, dim=-1)
     transcription = tokenizer.batch_decode(predicted_ids)
@@ -104,4 +107,4 @@ print("WER:", wer(result["text"], result["transcription"]))
 
 | "clean" | "other" |
 |---|---|
-| 2.2 | 5.2 |
+| 1.9 | 3.9 |