chausner-audeering
commented
2 years ago The theoretical value ranges of most features are typically not very useful as you would usually see very different ranges in practice on datasets. The value range of the delta features can be derived from the value range of the original feature. For instance, if the original feature range is a..b, the delta feature will have the (theoretical) range of (a-b)..(b-a).
I saw the closed issue asking about feature ranges and the suggestion to normalize data relative to all audio file outputs. Due to a relatively small dataset, I would like to normalize using the actual range of these features. Some I can find online, but I was wondering if there was anything in your docs about ranges of these features:
I know MFCCs are 0-1, but am especially having trouble finding ranges for the differential values of these coefficients...
pcm_RMSenergy_sma; pcm_fftMag_mfcc_sma[1]; pcm_fftMag_mfcc_sma[2]; pcm_fftMag_mfcc_sma[3]; pcm_fftMag_mfcc_sma[4]; pcm_fftMag_mfcc_sma[5]; pcm_fftMag_mfcc_sma[6]; pcm_fftMag_mfcc_sma[7]; pcm_fftMag_mfcc_sma[8]; pcm_fftMag_mfcc_sma[9]; pcm_fftMag_mfcc_sma[10]; pcm_fftMag_mfcc_sma[11]; pcm_fftMag_mfcc_sma[12]; pcm_zcr_sma; voiceProb_sma; F0_sma; pcm_RMSenergy_sma_de; pcm_fftMag_mfcc_sma_de[1]; pcm_fftMag_mfcc_sma_de[2]; pcm_fftMag_mfcc_sma_de[3]; pcm_fftMag_mfcc_sma_de[4]; pcm_fftMag_mfcc_sma_de[5]; pcm_fftMag_mfcc_sma_de[6]; pcm_fftMag_mfcc_sma_de[7]; pcm_fftMag_mfcc_sma_de[8]; pcm_fftMag_mfcc_sma_de[9]; pcm_fftMag_mfcc_sma_de[10]; pcm_fftMag_mfcc_sma_de[11]; pcm_fftMag_mfcc_sma_de[12]; pcm_zcr_sma_de; voiceProb_sma_de; F0_sma_de