Hello. I followed the official documentation to run decoding with this model. On the LibriSpeech test-clean set, I obtained WERs of only 2.2% for streaming and 1.83% for non-streaming, falling short of the official reported results of 1.63% and 1.78% respectively. Could you advise which detailed configurations I might have missed?
Hello. I followed the official documentation to run decoding with this model. On the LibriSpeech test-clean set, I obtained WERs of only 2.2% for streaming and 1.83% for non-streaming, falling short of the official reported results of 1.63% and 1.78% respectively. Could you advise which detailed configurations I might have missed?