[`Whisper`] Nit converting the tokenizer (#27349)

* `nospeech` instead of `nocaption` for the no speech token

* oups

src/transformers/models/whisper/convert_openai_to_hf.py

@@ -235,7 +235,7 @@ def convert_tiktoken_to_hf(
         "<|transcribe|>",
         "<|startoflm|>",
         "<|startofprev|>",
-        "<|nocaptions|>",
+        "<|nospeech|>",
         "<|notimestamps|>",
     ]
     # these are special tokens, not normalized