登入選單
返回Google圖書搜尋
Performance of the IBM Large Vocabulary Continuous Speech Recognition System on the ARPA Wall Street Journal Task
註釋Abstract: "This paper presents the results obtained when evaluating the IBM Research large vocabulary continuous speech recognition system on the Phase I component of the ARPA Wall Street Journal task. All experiments are conducted using statistical trigram language models with an underlying vocabulary of 20,000 words. The recognition system is evaluated on both development and test sets from the Phase I data, in the case of non-verbalized punctuation. Speaker-dependent, longitudinal speaker-dependent, and speaker-independent results are reported."