論文: EESEN:END-TO-END SPEECH RECOGNITION USING DEEP RNN MODELS AND WFST-BASED DECODING ...
論文: IMPROVING LATENCY CONTROLLED BLSTM ACOUSTIC MODELS FOR ONLINE SPEECH RECOGNITION 思想: BLSTM作為當前主流的序列建模算法,在語音識別領域取得了不錯的效果。但因為BLSTM的雙向LSTM結構,在序列建模時需要用到未來的時序信息,這使得算法在流式語音識別中受到制約,不滿足流式語音識別對輸出延遲的要求 而LC ...
2020-09-16 10:11 0 589 推薦指數:
論文: EESEN:END-TO-END SPEECH RECOGNITION USING DEEP RNN MODELS AND WFST-BASED DECODING ...
論文: CTC:Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks 思想: 語音識別中,一般包含語音 ...
論文: Deep-FSMN for Large Vocabulary Continuous Speech Recognition 思想: 對於大詞匯量語音識別,往往需要更深的網絡結構,但是當FSMN[1]或cFSMN[2]的結構很深時容易引發剃度消失和爆炸問題 ...
論文: SPEECH-TRANSFORMER: A NO-RECURRENCE SEQUENCE-TO-SEQUENCE MODELFOR SPEECH RECOGNITION ...
LAS: listen, attented and spell,Google 思想: sequence to sequence的思想,模型分為encoder和dec ...
論文: TRANSFORMER TRANSDUCER: A STREAMABLE SPEECH RECOGNITION MODELWITH TRANSFORMER ENCODERS A ...
論文: RNNT:SPEECH RECOGNITION WITH DEEP RECURRENT NEURAL NETWORKS,2013 LSTM結構: ...
論文: EXPLORING ARCHITECTURES, DATA AND UNITS FOR STREAMING END-TO-END SPEECH RECOGNITION WITH RNN- ...