Sequence discriminative training
語音識別聲學模型DNN訓練通常用cross-entropy（CE）作爲損失函數進行訓練，CE可以看做是KL散度的一種形式 ，用來評價期望分佈和當前訓練模型概率分佈的差距。方便計算，所以常常被用來作爲損失函數廣泛使用。然而基於幀識別的語音識別中，往往用WER或者CER，PER來作爲評價語音識別的準確率。損失函數和訓練目標不一致，因此[1]提出了基於序列區分度（Sequence Disc
相關文章
- 1. Self-critical Sequence Training
- 2. discriminative training鑑別性訓練
- 3. BART: Denoising Sequence-to-Sequence Pre-training翻譯
- 4. Self-critical Sequence Training for Image Captioning
- 5. 文獻閱讀筆記-MASS： Masked Sequence to Sequence Pre-training for Language Generation
- 6. 2018 Multi-University Training Contest 7 （hdu6395 Sequence）
- 7. 2018 Multi-University Training Contest 1 G（Chiaki Sequence Revisited）
- 8. Discriminative Correlation Filters (DCF)
- 9. Write a Sequence to Sequence (seq2seq) Model
- 10. Discriminative Transfer Learning for General Image Restoration
- 更多相關文章...
- • XML Schema sequence 元素 - XML Schema 教程
- • XSL-FO page-sequence-master 對象 - XSL-FO 教程
- • YAML 入門教程
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. Self-critical Sequence Training
- 2. discriminative training鑑別性訓練
- 3. BART: Denoising Sequence-to-Sequence Pre-training翻譯
- 4. Self-critical Sequence Training for Image Captioning
- 5. 文獻閱讀筆記-MASS： Masked Sequence to Sequence Pre-training for Language Generation
- 6. 2018 Multi-University Training Contest 7 （hdu6395 Sequence）
- 7. 2018 Multi-University Training Contest 1 G（Chiaki Sequence Revisited）
- 8. Discriminative Correlation Filters (DCF)
- 9. Write a Sequence to Sequence (seq2seq) Model
- 10. Discriminative Transfer Learning for General Image Restoration