Upload README.md with huggingface_hub

README.md

@@ -41,9 +41,19 @@ inputs = processor(waveform.squeeze().numpy(), sampling_rate=16000, return_tenso
 with torch.no_grad():
     logits = model(**inputs).logits
 
-pred_ids = torch.argmax(logits, dim=-1)[0]
-phone_seq = processor.tokenizer.decode(pred_ids, skip_special_tokens=True)
-print(phone_seq)
+pred_ids = torch.argmax(logits, dim=-1)[0].tolist()
+blank_id = processor.tokenizer.pad_token_id
+
+phone_tokens = []
+prev = None
+for token_id in pred_ids:
+    if token_id != blank_id and token_id != prev:
+        token = model.config.id2label.get(token_id, processor.tokenizer.convert_ids_to_tokens(token_id))
+        if token not in {"<PAD>", "<UNK>", "<BOS>", "<EOS>", "|"}:
+            phone_tokens.append(token)
+    prev = token_id
+
+print(" ".join(phone_tokens))
 ```
 
 ## Citation