Paper seminar by Odongo Steven Eyobu. Presented on: 2017-11-16
Title: Human activity recognition with smartphone sensors using deep learning neural networks
Authors: Charissa Ann Ronao, Sung-Bae Cho.
Journal publisher: Expert Systems with Applications
The article proposes:
The use of deep convolutional networks (convnets) for efficient and effective human activity recognition using smartphone inertial measurement unit (IMU) sensor data.
a. They claim that convnets provide an efficient way to automatically and adaptively extract robust features from raw sensor data.
b. By varying different temporal correlation sizes and pooling sizes, optimal dimensions of both parameters can be found through a greedy-wise search manner.
c. The final results in the article shows that the convent can achieve upto 94.79 accuracy on raw sensor test data and 95.75% with additional information of temporal fast fourier transform of the HAR datasert.
Paper seminar by Young Joon Moon. Presented on: 2017-11-16
a. Title: RASTA-PLP Speech Analysis
Authors: Hynek Hermansky et al.
b.Title: Revising Perceptual Linear Prediction(PLP)
Authors: Florian Honig et al.
Using those two papers, the concept of PLP and MFCC (Mel Frequency Cepstral Coefficients) was introduced and also the computation steps of PLP and MFCC were introduced. Based on those information, possible enhancement of performance of the filter banks, which are used those speech analysis methods, was discussed.
PLP has good performance in the present of noise when it is compared to MFCC. However, in a clean acoustic environment, both feature types have about equal results especially on HUB4 which is acoustically less homogeneous.