Sequence-to-Sequence Speech Recognition with Time-Depth Separable Convolutions
1.論文摘要 提出了一種time-depth separable 的卷積網絡結構，作爲ED模型的encoder,在顯著減少了參數量的同時增加了計算速度，並且可以維持較大的感受野範圍，在noisy LibriSpeech test set 取得了WER 22%的提升。 2.模型結構 encoder TDS 的卷積結構，採用了一個2d卷積，這裏輸入維度爲（batch_size, 1, time_ste
相關文章
- 1. 00036-Xception:Deep Learning with Depthwise Separable Convolutions
- 2. 翻譯Xception:Deep Learning with Depthwise Separable Convolutions
- 3. Inception系列4_Xception： Deep Learning with Depthwise Separable Convolutions
- 4. Food Log with Speech Recognition and NLP
- 5. (7) [IEEEAcoustic13] SPEECH RECOGNITION WITH DEEP RECURRENT NEURAL NETWORKS
- 6. 文章：Emotion Recognition From Speech With Recurrent Neural Networks
- 7. TRANSFORMER-TRANSDUCER:END-TO-END SPEECH RECOGNITION WITH SELF-ATTENTION
- 8. Voice Conversion by Cascading Automatic Speech Recognition and Text-to-Speech Synthesis with Prosody
- 9. speech recognition resource
- 10. Xception：Deep Learning with Depthwise Separable Convolutions論文閱讀解釋
- 更多相關文章...
- • XSLT
- • XSL-FO page-number 對象 - XSL-FO 教程
- • 爲了進字節跳動，我精選了29道Java經典算法題，帶詳細講解
- • 算法總結-股票買賣
相關標籤/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
歡迎關注本站公眾號,獲取更多信息
相關文章
- 1. 00036-Xception:Deep Learning with Depthwise Separable Convolutions
- 2. 翻譯Xception:Deep Learning with Depthwise Separable Convolutions
- 3. Inception系列4_Xception： Deep Learning with Depthwise Separable Convolutions
- 4. Food Log with Speech Recognition and NLP
- 5. (7) [IEEEAcoustic13] SPEECH RECOGNITION WITH DEEP RECURRENT NEURAL NETWORKS
- 6. 文章：Emotion Recognition From Speech With Recurrent Neural Networks
- 7. TRANSFORMER-TRANSDUCER:END-TO-END SPEECH RECOGNITION WITH SELF-ATTENTION
- 8. Voice Conversion by Cascading Automatic Speech Recognition and Text-to-Speech Synthesis with Prosody
- 9. speech recognition resource
- 10. Xception：Deep Learning with Depthwise Separable Convolutions論文閱讀解釋