Factorized Hidden Variability Learning For Adaptation Of Short Duration Language Identification Models

基于因子分解的隐层变量学习,应用于短语句语种识别模型的自适应

LFVs(Language Feature Vectors,语种特征向量)[11],与BSVs(Bottleneck Speaker Vectors)类似,即瓶颈特征

3.1. 神经元调制

由于说话人特性的变化反映在语音信号中,因此将表示说话人适应声学特性的特征拼接到特征中。如VTLN或fMLLR,是直接对声学特征进行操作的自适应方法。可以训练一个说话人自适应系统以基于说话人属性直接对输入特征进行转换,这样效果与基于i-Vector的自适应类似[8]。但是与说话人变化特性相比,语言特性是更高阶的概念。在某些方面基于声学。例如,具有相同音素的多种语言,可以在某种程度上可以观察到语言特定属性。但是,声学特征变换适应性方法无法考虑到音位配列学或者不同声学单元集的知识。在这里,在更深层次的DNN处添加特征可能会改善自适应性。[17]基于Meta-PI网络进行了尝试。关键点是使用Meta-PI连接,它允许通过将隐层单元乘以系数来调制神经元的输出。应用于语种自适应,我们用LFV来对隐藏层的输出进行调制。基于语种特征的调制,LSTM单元的输出被衰减或增强。这迫使隐藏层中的单元基于语种特征来学习或适应。调制可以被认为与Dropout有关[18],其中网络连接以随机概率被丢弃。在结果部分中,我们将此方法称为"LFV调制"。

我们使用了如图1所示的网络配置。基??本架构受百度Deepspeech 2的启发。它将两个TDNN/CNN层与4个双向LSTM层组合在一起。输出层是一个前馈层,它将最后一个LSTM层的输出映射到目标。将每层LSTM单元维数设定为LFV维数的数倍。这样就可以构建包含相同单位数量的LSTM单元的隐藏层组。然后用LFV的某一维对每组的输出进行调制。该图显示了两种配置,"LFV 拼接"和"LFV 调制",但一次只应用一种方法。在初步实验中,我们得出在第二个LSTM层的输出处进行调制可以获得最佳性能。

原文地址:https://www.cnblogs.com/JarvanWang/p/9152674.html

时间: 2024-10-01 00:43:29

Factorized Hidden Variability Learning For Adaptation Of Short Duration Language Identification Models的相关文章

关于Increased rates of convergence through learning rate adaptation一文的理解

原文地址:http://www.researchgate.net/profile/Robert_Jacobs9/publication/223108796_Increased_rates_of_convergence_through_learning_rate_adaptation/links/0deec525d8f8dd5ade000000.pdf 已经看了CNN,rbm,sae等网络及算法,所有网络在训练时都需要一个learning rate,一直以来都觉得这个量设为定值即可,现在才发现其实

Machine and Deep Learning with Python

Machine and Deep Learning with Python Education Tutorials and courses Supervised learning superstitions cheat sheet Introduction to Deep Learning with Python How to implement a neural network How to build and run your first deep learning network Neur

Awesome Machine Learning

Awesome Machine Learning  A curated list of awesome machine learning frameworks, libraries and software (by language). Inspired by awesome-php. If you want to contribute to this list (please do), send me a pull request or contact me @josephmisiti Als

Deep Learning in a Nutshell: History and Training

Deep Learning in a Nutshell: History and Training This series of blog posts aims to provide an intuitive and gentle introduction to deep learning that does not rely heavily on math or theoretical constructs. The first part in this series provided an

(转)Understanding, generalisation, and transfer learning in deep neural networks

Understanding, generalisation, and transfer learning in deep neural networks FEBRUARY 27, 2017 This is the first in a series of posts looking at the 'top 100 awesome deep learning papers.' Deviating from the normal one-paper-per-day format, I'll take

Machine Learning for Developers

Machine Learning for Developers Most developers these days have heard of machine learning, but when trying to find an 'easy' way into this technique, most people find themselves getting scared off by the abstractness of the concept of Machine Learnin

[C3] Andrew Ng - Neural Networks and Deep Learning

About this Course If you want to break into cutting-edge AI, this course will help you do so. Deep learning engineers are highly sought after, and mastering deep learning will give you numerous new career opportunities. Deep learning is also a new "s

NTU-Coursera机器学习:机器学习基石 (Machine Learning Foundations)

课讲内容 这门课以8周设计,分成 4个核心问题,每个核心问题约需2周的时间来探讨.每个约2个小时的录影中,每个小时为一个主题,以会各分成4到5个小段落,每个段落里会有一个后多个随堂的练习.我们在探讨每个核心问题的第二周.依上所述,課程的規畫如下: When Can Machines Learn? [何时可以使用机器学习] 第一周:(NTU-Coursera机器学习:机器学习问题与二元分类) 第一讲:The Learning Problem [机器学习问题]第二讲:Learning to Answ

(转) Learning Deep Learning with Keras

Learning Deep Learning with Keras Piotr Migda? - blog Projects Articles Publications Resume About Photos Learning Deep Learning with Keras 30 Apr 2017 ? Piotr Migda? ? [machine-learning] [deep-learning] [overview] I teach deep learning both for a liv