路喵甲自學紀錄：AI 與其他有的沒的

Taiwanese-Corpus https://github.com/Taiwanese-Corpus 意傳.台灣 https://github.com/i3thuan5 其中這個專案蠻完整的，只是頁面 demo 似乎掛了... https://github.com/i3thuan5/tai5-uan5_gian5-gi2_hok8-bu7/wiki 中研院 語料庫資源 http://elearning.ling.sinica.edu.tw/resources.html librivox 中文語料 https://librivox.org/search?primary_key=14&search_category=language&search_page=1&search_form=get_results Dysarthric speech database for universal access research https://experts.illinois.edu/en/publications/dysarthric-speech-database-for-universal-access-research Kaggle: Mozilla https://www.kaggle.com/mozillaorg/datasets LibriVox https://librivox.org/ https://zh.wikipedia.org/wiki/LibriVox The TORGO Database: Acoustic and articulatory speech from speakers with dysarthria http://www.cs.toronto.edu/~complingweb/data/TORGO/torgo.html

感覺沒有科大訊飛的辨識來的準確（中文） Cloud Speech-to-Text https://cloud.google.com/speech-to-text/ python-client-migration https://cloud.google.com/speech-to-text/docs/python-client-migration Language Support https://cloud.google.com/speech-to-text/docs/languages

什么是迁移学习 (Transfer Learning)？这个领域历史发展前景如何？ https://www.zhihu.com/question/41979241 迁移学习 Transfer Learning https://morvanzhou.github.io/tutorials/machine-learning/tensorflow/5-16-transfer-learning/ Automatic Speaker Recognition using Transfer Learning https://towardsdatascience.com/automatic-speaker-recognition-using-transfer-learning-6fab63e34e74 transferlearning Everything about Transfer Learning and Domain Adaptation--迁移学习 http://transferlearning.xyz/ https://github.com/jindongwang/transferlearning 《小王爱迁移》系列之十三：在线迁移学习 https://zhuanlan.zhihu.com/p/33557802 2016《A survey of transfer learning》迁移学习笔记 https://blog.csdn.net/LiGuang923/article/details/81948693 初探异构迁移学习 https://blog.csdn.net/qq_34564612/article/details/79193050

ML Lecture 9: Tips for Training DNN http://violin-tao.blogspot.com/2017/07/ml-tips-for-training-dnn.html 以100張圖理解 Neural Network -- 觀念與實踐 這篇很有趣 https://ithelp.ithome.com.tw/users/20001976/ironman/1395 兼容 backend https://morvanzhou.github.io/tutorials/machine-learning/keras/1-3-backend/ How to get the number of elements in a list in Python? https://stackoverflow.com/questions/1712227/how-to-get-the-number-of-elements-in-a-list-in-python Python中的numpy矩阵运算 https://blog.csdn.net/lilong117194/article/details/78308780 Merge two different deep learning models in Keras 一樣是序列不是平行方式 https://stackoverflow.com/questions/47035367/merge-two-different-deep-learning-models-in-keras

選自arXiv 本論文技術性地介紹了三種最常見的神經網絡：前饋神經網絡、卷積神經網絡和循環神經網絡。且該文詳細介紹了每一種網絡的基本構建塊，其包括了基本架構、傳播方式、連接方式、激活函數、反向傳播的應用和各種優化算法的原理。本文不僅介紹了這三種神經網絡的基本原理與概念，同時還用數學表達式正式地定義了這些概念。這是一份十分全面的神經網絡綜述論文。 http://bangqu.com/53vxzK.html

A TensorFlow Implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model https://github.com/Kyubyong/tacotron

Speech recognition using the tensorflow deep learning framework, sequence-to-sequence neural networks https://github.com/pannous/tensorflow-speech-recognition

Front-end speech processing aims at extracting proper features from short- term segments of a speech utterance, known as frames. It is a pre-requisite step toward any pattern recognition problem employing speech or audio (e.g., music). Here, we are interesting in voice disorder classification. That is, to develop two-class classifiers, which can… https://github.com/gionanide/Speech_Signal_Processing_and_Classification https://github.com/gionanide/Speech_Signal_Processing_and_Classification/blob/master/Short_Programming_Project/university-groningen-short.pdf

使用Keras中的RNN模型进行时间序列预测 http://resuly.me/2017/08/16/keras-rnn-tutorial/ Docs » Layers » 循环层 Recurrent https://keras.io/zh/layers/recurrent/ Keras之LSTM源码阅读笔记 https://blog.csdn.net/silent56_th/article/details/73442391 [Keras] 利用Keras建構LSTM模型，以Stock Prediction 為例 1 https://medium.com/@daniel820710/%E5%88%A9%E7%94%A8keras%E5%BB%BA%E6%A7%8Blstm%E6%A8%A1%E5%9E%8B-%E4%BB%A5stock-prediction-%E7%82%BA%E4%BE%8B-1-67456e0a0b 人人都能看懂的GRU https://zhuanlan.zhihu.com/p/32481747 从 RNN, LSTM, GRU 到 SRU https://mp.weixin.qq.com/s/WSNb7VLdQQHBmRipuFEZOA? CNN（卷積神經網絡）、RNN（循環神經網絡）、DNN（深度神經網絡）的內部網絡結構的區別 https://kknews.cc/zh-tw/news/6n84mm3.html [資料分析&機器學習] 第5.1講: 卷積神經網絡介紹(Convolutional Neural Network) https://medium.com/jameslearningnote/%E8%B3%87%E6%96%99%E5%88%86%E6%9E%90-%E6%A9%9F%E5%99%A8%E5%AD%B8%E7%BF%92-%E7%AC%AC5-1%E8%AC%9B-%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E7%B5%A1%E4%BB%8B%E7%B4%B9-convolutional-neural-network-4f8249d65d4f 卷积神经网络（CNN）入门讲解 https://zhuanlan.zhi...

你必须要知道CNN模型：ResNet https://zhuanlan.zhihu.com/p/31852747 ResNet的理解及其Keras实现 http://lanbing510.info/2017/08/21/ResNet-Keras.html Docs » Applications https://keras.io/applications/ Keras 入门课4 -- 使用ResNet识别cifar10数据集 https://blog.csdn.net/tsyccnh/article/details/78865167 無需數學背景，讀懂ResNet、Inception和Xception三大變革性架構 http://bangqu.com/4Zy3e7.html keras系列︱Application中五款已训练模型、VGG16框架（Sequential式、Model式）解读（二） https://hk.saowen.com/a/fd5f65fb61d1d7045a9ae8415f61ba77750f3532022d500925b02896700762cf 學習使用Keras建立卷積神經網路 https://chtseng.wordpress.com/2017/09/23/%E5%AD%B8%E7%BF%92%E4%BD%BF%E7%94%A8keras%E5%BB%BA%E7%AB%8B%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E8%B7%AF/ deep-learning-models https://github.com/fchollet/deep-learning-models Docs » Other » Application应用 https://keras-cn.readthedocs.io/en/latest/other/application/#resnet50

语音合成vocoder（三） spectral envelope参数 https://blog.csdn.net/xmdxcsj/article/details/72419948 STRAIGHT This is a speech analysis, modification and synthesis system https://github.com/shuaijiang/STRAIGHT Sprocket Voice Conversion Tool Kit https://github.com/k2kobayashi/sprocket https://nuss.nagoya-u.ac.jp/s/h8YKnq6qxjjxtU3 WaveNet: A Generative Model for Raw Audio https://deepmind.com/blog/wavenet-generative-model-raw-audio/ WaveNet初步理解 https://www.cnblogs.com/zhanxiage1994/p/7872101.html wavenet_vocoder https://github.com/r9y9/wavenet_vocoder tensorflow_wavenet_vocoder https://github.com/azraelkuan/tensorflow_wavenet_vocoder

Docs » Layers » 池化层 Pooling https://keras.io/zh/layers/pooling/ https://keras.io/zh/layers/pooling/#maxpooling3d Docs » Layers » 核心网络层 https://keras.io/zh/layers/core/ Docs » Layers » 卷积层 Convolutional https://keras.io/zh/layers/convolutional/ Docs » 初始化 Initializers https://keras.io/zh/initializers/ Docs » 正则化 Regularizers https://keras.io/zh/regularizers/ Docs » Layers » 标准化层 Normalization https://keras.io/zh/layers/normalization/ Docs » 损失函数 Losses https://keras.io/zh/losses/#categorical_crossentropy Docs » Callbacks https://keras.io/callbacks/ Docs » 模型 » 函数式 API https://keras.io/zh/models/model/#model Docs » 工具 https://keras.io/zh/utils/ 如何获取中间层的输出？ https://keras-cn.readthedocs.io/en/latest/for_beginners/FAQ/

聊一聊深度学习的activation function https://zhuanlan.zhihu.com/p/25110450 高级激活层 Advanced Activations https://keras.io/zh/layers/advanced-activations/ How could we use Leaky ReLU and Parametric ReLU as activation function ? https://github.com/keras-team/keras/issues/117 26种神经网络激活函数可视化 這篇分析的很棒 https://www.jiqizhixin.com/articles/2017-10-10-3 激活函数ReLU、Leaky ReLU、PReLU和RReLU https://blog.csdn.net/qq_23304241/article/details/80300149 深度学习基础（十二）—— ReLU vs PReLU https://blog.csdn.net/lanchunhui/article/details/52644823

淺談降維方法中的 PCA 與 t-SNE https://medium.com/d-d-mag/%E6%B7%BA%E8%AB%87%E5%85%A9%E7%A8%AE%E9%99%8D%E7%B6%AD%E6%96%B9%E6%B3%95-pca-%E8%88%87-t-sne-d4254916925b 資料科學/機器學習的好用入門工具 t-SNE 幫你看見高維度數值資料 https://newtoypia.blogspot.com/2017/07/t-sne.html MATLAB: t-SNE https://www.mathworks.com/help/stats/t-sne.html 从SNE到t-SNE再到LargeVis http://bindog.github.io/blog/2016/06/04/from-sne-to-tsne-to-largevis/ scikit-learn https://scikit-learn.org/stable/ sklearn.manifold.TSNE https://scikit-learn.org/stable/modules/generated/sklearn.manifold.TSNE.html

https://ithelp.ithome.com.tw/articles/10192162 https://keras.io/applications/ 來測試一下 VGG16/19 跟 ResNet50...

教電腦畫畫：初心者的生成式對抗網路(GAN)入門筆記（TensorFlow + Python3） https://medium.com/@gau820827/%E6%95%99%E9%9B%BB%E8%85%A6%E7%95%AB%E7%95%AB-%E5%88%9D%E5%BF%83%E8%80%85%E7%9A%84%E7%94%9F%E6%88%90%E5%BC%8F%E5%B0%8D%E6%8A%97%E7%B6%B2%E8%B7%AF-gan-%E5%85%A5%E9%96%80%E7%AD%86%E8%A8%98-tensorflow-python3-dfad71662952 這樣的解說還蠻有趣的

但這樣的架構也不是沒有極限的，作者提到了只有在兩個domain分享類似的語義內容才能成功，斑馬跟馬之間可以成功轉換，但狗跟貓之間就無法轉換 https://data-sci.info/2017/05/25/cyclegan/ https://github.com/junyanz/CycleGAN https://hardikbansal.github.io/CycleGANBlog/ https://github.com/Shaofanl/CycleGAN-keras

https://towardsdatascience.com/how-i-implemented-iphone-xs-faceid-using-deep-learning-in-python-d5dbaa128e1d code: https://github.com/normandipalo/faceID_beta 該弄一隻來測試了？

http://technews.tw/2018/04/23/image-reconstruction-by-deep-learning/

https://read01.com/zh-tw/MaOn8B.html#.Wr9fXohubIU 比較一下