China
Africa
Explore chapters and articles related to this topic
Radar Monitoring of Humans with Assistive Walking Devices
Published in Moeness G. Amin, Radar for Indoor Monitoring, 2017
Ann-Kathrin Seifert, Moeness G. Amin, Abdelhak M. Zoubir
The work by Gürbüz et al. (2016) investigates the effect of different walking aids on the characteristics of radar return time–frequency signatures of human walk. Different degrees of mobility are compared: a normal, that is, unaided, walk, walking with a limp, walking with a cane or tripod, walking with a walker, and using a wheelchair. Besides using different radar systems (ultra high frequency (UHF), L-band, K-band, 24/77 GHz), ultrasound systems (20–80 kHz) are examined. Micro-Doppler features are obtained from the most commonly used TFR, the spectrogram. Machine learning and pattern recognition techniques are applied to classify the different walking styles. Different types of micro-Doppler features are investigated. Traditionally, physical features are extracted from the spectrogram and relate to, for example, the mean gait velocity, mean stride frequency, stride variability, and bandwidth of Doppler signal. Moreover, features are extracted by making use of speech processing algorithms such as linear prediction coding (LPC) coefficients, cepstrum coefficients, and mel-frequency cepstrum coefficients (MFCCs). Feature selection is performed utilizing the mutual information feature selector under uniform information distribution algorithm.
Analysis of RNNs and Different ML and DL Classifiers on Speech-Based Emotion Recognition System Using Linear and Nonlinear Features
Published in Amit Kumar Tyagi, Ajith Abraham, Recurrent Neural Networks, 2023
Shivesh Jha, Sanay Shah, Raj Ghamsani, Preet Sanghavi, Narendra M. Shekokar
MFC represents short-term power spectrum of sound. It is basically a linear cosine transform of a log power spectrum on a nonlinear mel scale of frequency. Mel-frequency cepstral coefficients (MFCCs) are the coefficients that are building blocks of Mel-frequency cepstral coefficient (MFC). They are a derivative of a type of audio clip cepstral representation. The main distinction between the cepstrum and the mel-frequency cepstrum is that the MFC’s frequency bands are uniformly spaced on the mel scale, which more closely approximates the human auditory system’s response than the normal spectrum’s linearly spaced frequency bands. This frequency warping leads to a better representation of sound—for instance, in audio compression.
Fundamentals of Speech Processing
Published in Shaila Dinkar Apte, Random Signal Processing, 2017
The mel-frequency cepstrum (MFC) can be defined as the short time power spectrum of speech signal, which is calculated as a linear cosine transform of a log power spectrum on a nonlinear mel scale of frequency. MFCC are the coefficients evaluated in the MFC representation. Let us see the difference between the cepstrum and the mel cepstrum. In the case of MFC, the frequency bands are essentially equally spaced on the mel scale. This mel scale approximates the human auditory system’s response more closely than the linearly spaced frequency bands. The mel scale warps the frequency and allows better representation that is similar to human auditory system.
Design of a speech-enabled 3D marine compass simulation system
Published in Ships and Offshore Structures, 2018
Bin Fu, Hongxiang Ren, Jingjing Liu, Xiaoxi Zhang
Because the human auditory system is nonlinear, its sensitivity to sound and audio frequencies are out of proportion. Below 1000 Hz, the Mel frequency tends toward a linear distribution, while above 1000 Hz, it tends toward a logarithmic distribution. The frequency band of the Mel frequency cepstrum is equidistantly partitioned on the Mel scale, which is closer to the human auditory system than any other. Therefore, MFCC can solve the problem of inconsistent frequency distribution. The relationship between Mel frequency and linear frequency is calculated as follows: where Mel(f) denotes the Mel frequency, and f is the linear frequency. Calculating the MFCC parameter involves two main stages: framing and filter analysis. The calculation process is shown in Figure 9.
Multiple Oracle consensus for weakly supervised defect detection in concrete structures using audio data
Published in Advanced Robotics, 2021
Jun Younes Louhi Kasahara, Atsushi Yamashita, Hajime Asama
Mel-Frequency Cepstrum Coefficients (MFCC) [21, 22] are a popular audio feature vector made to mimic the human audition. Since the hammering test is traditionally conducted by human inspectors, it can be expected that the human auditory perception is a good feature space for discriminating defect hammering samples. Therefore, MFCC were used in [23] and proved to be a suitable audio feature representation for defect hammering sample discrimination. Therefore, MFCC feature vectors are built using the normalized Fourier spectrum and used as our initial feature vectors. Since MFCC are hand-crafted feature vectors, it can be expected that better feature vectors for the task at hand exist: MFCC are simply used here are initial feature space to be fine-tuned by RCA.
Wood material recognition for industrial applications
Published in Systems Science & Control Engineering, 2018
Haibin Fu, Huaping Liu, Xiaoyan Deng, Fuchun Sun
The traditional Mel frequency cepstrum coefficient assumes that the sound signal is short and stable, obtained with a fixed window Fourier transform. From the principle of uncertainty, this assumption makes the details of the spectrum of the sound fuzzy and loses certain information. The use of discrete wavelet Fourier transform to replace the fast Fourier transform in the extraction process of the traditional MFCC coefficients, can be better to solve this problem.