Posted on Categories:CS代写, Machine Learning, 机器学习, 计算机代写

# 计算机代写|深度学习代写Deep Learning代考|CS7643 Time-Dependency Model

avatest™

## avatest™帮您通过考试

avatest™的各个学科专家已帮了学生顺利通过达上千场考试。我们保证您快速准时完成各时长和类型的考试，包括in class、take home、online、proctor。写手整理各样的资源来或按照您学校的资料教您，创造模拟试题，提供所有的问题例子，以保证您在真实考试中取得的通过率是85%以上。如果您有即将到来的每周、季考、期中或期末考试，我们都能帮助您！

•最快12小时交付

•200+ 英语母语导师

•70分以下全额退款

## 计算机代写|深度学习代写Deep Learning代考|Time-Dependency Model

In this subsection, different time-dependency models are compared. Figure 3.8a shows the results for the model with CNN framewise modelling and attentionpooling and three different time-dependency models. The model on the very lefthand side is again the CNN-SA-AP model with self-attention for time-dependency modelling, the CNN-LSTM-AP model uses a BiLSTM network with one layer and 128 hidden units, and the CNN-Skip-AP model skips the time-dependency stage and directly pools the $\mathrm{CNN}$ outputs. It can be seen that difference between the different time-dependency models is not as large as for the framewise models. That being said, it can be noted that the inclusion of a time-dependency model improves the overall performance, where the self-attention network shows to outperform the LSTM network. The PCC of the self-attention network is on average around 0.15 higher than the $\mathrm{PCC}$ of the model without time dependency.

The different time-dependency models can also be combined by applying an LSTM network with the following self-attention network or vice versa. The results of these combinations are shown in Fig. $3.8 \mathrm{~b}$, where it can be seen that both combinations perform on average worse than using only an LSTM or SA network. Overall, it can be concluded that it is beneficial for a speech quality model to include a time-dependency stage. The difference between using a self-attention or LSTM network for this purpose is only small, but the SA network showed to give the best results on average.

## 计算机代写|深度学习代写Deep Learning代考|Pooling Model

In this subsection different pooling models are compared, where for framewise modelling a CNN is applied. The different pooling mechanisms are average-pooling (Avg), max-pooling (Max), and attention-pooling. In the case of the recurrent LSTM network, also the output of the last time step can be used for pooling (Last).

Figure $3.9 \mathrm{~b}$ in the middle compares the proposed CNN-SA-AP model to CNNSA models with max- or average-pooling. Although the performance difference is relatively small, it can be seen that attention-pooling outperforms the other two methods, where average-pooling and max-pooling achieve approximately the same results. In the case of using an LSTM for time-dependency modelling in Fig. 3.9c, attention-pooling again outperforms the other mechanisms, where average-pooling achieves similar results. Last-step-pooling and max-pooling are outperformed by the other methods. In contrast to that, in Fig. 3.9a, where the output of the CNN is directly pooled without TD model, max-pooling outperforms average pooling. Overall, again attention-pooling achieves the best results. Interestingly, in the case that no TD model is applied, max-pooling gives better results than average-pooling, which is in line with the quality perception of speech communication users that tend to give more weight to segments in the speech file with poor quality. However, this out-weighting of poor-quality segments seems to be modelled by the timedependency stage, and therefore average-pooling achieves similar or better results than max-pooling if a TD stage is included. Overall, it can be concluded that the choice of the pooling stage only has a small influence on the performance if selfattention is applied and that attention-pooling achieves on average the best result for all tested cases.

## MATLAB代写

MATLAB 是一种用于技术计算的高性能语言。它将计算、可视化和编程集成在一个易于使用的环境中，其中问题和解决方案以熟悉的数学符号表示。典型用途包括：数学和计算算法开发建模、仿真和原型制作数据分析、探索和可视化科学和工程图形应用程序开发，包括图形用户界面构建MATLAB 是一个交互式系统，其基本数据元素是一个不需要维度的数组。这使您可以解决许多技术计算问题，尤其是那些具有矩阵和向量公式的问题，而只需用 C 或 Fortran 等标量非交互式语言编写程序所需的时间的一小部分。MATLAB 名称代表矩阵实验室。MATLAB 最初的编写目的是提供对由 LINPACK 和 EISPACK 项目开发的矩阵软件的轻松访问，这两个项目共同代表了矩阵计算软件的最新技术。MATLAB 经过多年的发展，得到了许多用户的投入。在大学环境中，它是数学、工程和科学入门和高级课程的标准教学工具。在工业领域，MATLAB 是高效研究、开发和分析的首选工具。MATLAB 具有一系列称为工具箱的特定于应用程序的解决方案。对于大多数 MATLAB 用户来说非常重要，工具箱允许您学习应用专业技术。工具箱是 MATLAB 函数（M 文件）的综合集合，可扩展 MATLAB 环境以解决特定类别的问题。可用工具箱的领域包括信号处理、控制系统、神经网络、模糊逻辑、小波、仿真等。