Skip to main content

Table 4 Significant works on infant cry detection

From: A review of infant cry analysis and classification

Literature Dataset Features Classifiers Performance
Chang [48] (2019) Self-recorded (crying with TV, speech, etc.) Spectrogram CNN 99.83%
Manikanta [25] (2019) Recorded in homes (crying with ac, fan, etc.) MFCC 1D-CNN, FFNN, SVM 86%
Dewi [64] (2019) Self-recorded (cry and not cry) LFCC KNN 90%
Gu [16] (2018) Self-recorded (crying with laughter, barking, etc.) LPC Dynamic time warping 97.1%
    algorithm  
Ferretti [18] (2018) Real Dataset: recorded in the NICU of a hospital; Log-Mel Coefficients CNN 86.58% on real dataset,
  Synthetic DB: crying with speech, “beep” sounds, etc.)    92.92% on synthetic
     dataset
Feier [87] (2017) TUT Rare Sound Events 2017 (crying with Log-amplitude mel-spectrogram CRNN 85% for baby crying detection,
  “glass breaking”, “gunshot”, etc.)    87% for all
     three targets
Torres [27] (2017) Online resources (crying with adult Voiced unvoiced counter, Consecutive F0 Support Vector Data AUC 92%
  cry, vacuum cleaning, etc.) and harmonic ratio accumulation,MFCC Description (SVDD),CNN  
Lavner [17] (2016) Recorded in domestic environment (crying MFCC, Pitch, Formants, etc. CNN 95%
  with speech, door opening, etc.)