A Novel Audio-Perception-Based Algorithm for Physiological Monitoring

Figures

• 

  Figure 1. System architecture of the audio separation-based monitoring system.

• 

  Figure 2. Comparison of spectrograms before and after filtering. (a) Raw spectrogram. (b) Spectrogram after band-pass filtering.

• 

  Figure 3. Autocorrelation analysis of audio signals. (a) Raw autocorrelation output. (b) Peaks identified by periodicity templates.

• 

  Figure 4. Audio separation into physiological components.

• 

  Figure 5. Heartbeat segmentation using short-term energy and zero-crossing rate. (Top): Audio waveform. (Bottom): Short-term energy curve of the same signal. Blue regions in both plots indicate segmented heartbeat intervals.

• 

  Figure 6. Schematic diagram of the CNN architecture for breathing classification.

• 

  Figure 7. BiLSTM architecture for respiratory signal analysis.

• 

  Figure 8. Cumulative distribution function (CDF) of respiratory rate errors in three environments. The x-axis represents the absolute error in respiratory rate (breaths per minute, bpm), and the y-axis indicates the proportion of samples with errors less than or equal to each x-axis value. The three environments—confined empty room, gym, and playground—represent increasing levels of background noise.

• 

  Figure 9. Incorrect respiratory rate in different environments.

• 

  Figure 10. ROC curves of RF, SVM, CNN, and GMM models in respiratory sound and noise detection.

• 

  Figure 11. Impact of experimental environments.

• 

  Figure 12. Impact of experimental environments on the mean absolute error of the respiratory rate.