What is the best feature extraction tool for (sound wave)? except for mfcc
If your question is about problems or situations related to perception of sound, then the features used should be related to that: psyco-acoustic measures such as loudness, sharpness and roughness might then be useful.
Perhaps something to have a look at is the application of scattering networks to classify audio/music signals [Andén et al. https://www.di.ens.fr/~mallat/papiers/IEEESignalAndenLostanlen.pdf]. The paper also bridges the gap between time-frequency analysis (e.g. MFCC, mel-spectrograms, etc) and deep neural networks.
I have experience with the Wigner Ville distribution, cepstral analysis, short-time Fourier transform and wavelet transform fed into a Convolutional Neural Network. There are literature sources that argue why WVD is superior to other time-frequency analysis transforms for non-stationary classification tasks with CNNs. In our application (classification of ultrasound backscatter), WVD, STFT and WT gave comparable results, while cepstral analysis was inferior.
Source: https://ieeexplore.ieee.org/document/8081222
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Extraction