I am doing thesis on baby cry prediction.
i had dataset of baby cries and non- baby cries of two classes.
i want use Mfcc feature extraction technique to identify important components of audio signal and train a model using this feature.
my question is how mfcc knows to select important components of signal.
i got a mfcc plot of baby cry.
can any one explain this plot?
MFCCs are the Mel Frequency Cepstral Coefficients. MFCC takes into account human perception for sensitivity at appropriate frequencies by converting the conventional frequency to Mel Scale, and are thus suitable for speech recognition tasks quite well (as they are suitable for understanding humans and the frequency at which humans speak/utter). I would suggest you to try to extract MFCCs from your audio signal. We generally take 12-13 Mel Frequency coefficients into consideration as features when training models. Other than that, I believe it would be useful if you consider other features as well. For your task on baby cry prediction, I would suggest you to use Volume, Energy, Pitch, Zero Crossing Rate, Spectral Centroid etc. as some additional features along with MFCC. I also suggest you to use some feature selection techniques like Principal Component Analysis (PCA) or t-sne (t-distributed stochastic neighbor embedding) for identifying optimal features. This would help you get better results.
MFCCs are the Mel Frequency Cepstral Coefficients. MFCC takes into account human perception for sensitivity at appropriate frequencies by converting the conventional frequency to Mel Scale, and are thus suitable for speech recognition tasks quite well (as they are suitable for understanding humans and the frequency at which humans speak/utter). I would suggest you to try to extract MFCCs from your audio signal. We generally take 12-13 Mel Frequency coefficients into consideration as features when training models. Other than that, I believe it would be useful if you consider other features as well. For your task on baby cry prediction, I would suggest you to use Volume, Energy, Pitch, Zero Crossing Rate, Spectral Centroid etc. as some additional features along with MFCC. I also suggest you to use some feature selection techniques like Principal Component Analysis (PCA) or t-sne (t-distributed stochastic neighbor embedding) for identifying optimal features. This would help you get better results.
Joyjit has a great answer and I agree with that.
I also would suggest checking the following URLs to learn more about MFCC:
https://medium.com/prathena/the-dummys-guide-to-mfcc-aceab2450fd
https://medium.com/@LeonFedden/comparative-audio-analysis-with-wavenet-mfccs-umap-t-sne-and-pca-cb8237bfce2f
Yes, Joyjit has explained this nicely. MFCCs are essentially like taking a Fourier Transform (or in your case, a spectrogram) of the signal, however, MFCCs use Mel scaling to try to model the way that the human hearing audiotory system perceives sounds, rather than describe them on a purely frequency (Hz) basis. This means that the MFCC should represent the textural or timbre of the signal (the baby cry) as we might hear it (e.g. a 'piercing' cry or a 'discontent' cry).
- Badges
- Science topic