Complexity analysis of normal and deaf infant cry acoustic waves

Kathiresan Manickam, Haizhou Li

TL;DR

This paper introduces a novel method for distinguishing between normal and deaf infant cries by analyzing the complexity of their acoustic waveforms using approximate entropy. The findings suggest that this complexity analysis can serve as a valuable tool for early identification of hearing impairment in infants.

What Problem Does This Paper Address?

Infant cries, while seemingly simple, are rich in information about an infant's physiological and psychological state. Different cry types (e.g., pain, hunger) are associated with distinct acoustic characteristics. However, identifying subtle differences, such as those between normal and deaf infants, can be challenging. While it's known that deaf infants may exhibit different vocalization patterns due to a lack of auditory feedback, a robust, objective method for distinguishing their cries from those of hearing infants is needed. This paper aims to address this gap by exploring the complexity of cry acoustic waves as a potential discriminant feature.

Methods

The core of this paper's methodology lies in the application of approximate entropy (ApEn) to quantify the complexity of infant cry waveforms. ApEn is a non-linear signal processing technique that measures the regularity of a time series. A higher ApEn value indicates a more complex and less predictable signal.

The authors first extract segments of infant cry waveforms. To normalize these signals for comparison across different infants, they compute the auto-covariance function of each frame. This function helps to reduce noise and highlight underlying patterns. The ApEn is then calculated from this auto-covariance function. The complexity of each frame is expressed as a single value, which is then aggregated (using the median) for each infant to provide a representative complexity measure.

Datasets

The study analyzed a corpus of infant crying samples from infants ranging from newborns up to 7 months old. This corpus consisted of:

  • 31 normal infant cries
  • 103 deaf infant cries

The cry signals were recorded and sampled at 8 KHz.

Key Insights and Ideas

  • Complexity as a Discriminator: The study reveals that the complexity of infant cry waveforms, as measured by approximate entropy, differs significantly between normal and deaf infants. This suggests that the acoustic characteristics of their cries are fundamentally different in terms of their underlying structure and predictability.
  • Auto-covariance for Noise Reduction: Using the auto-covariance function of cry segments proved effective in normalizing the signals and preparing them for complexity analysis, making the comparison between different infants more reliable.
  • Potential for Early Detection: The ability to distinguish between normal and deaf infant cries using this objective measure opens up possibilities for earlier identification of hearing impairment, which is crucial for timely intervention and improved developmental outcomes.

Key Results

The study found a distinct separation in the distribution of median complexity values between the deaf and normal infant populations. A Wilcoxon Rank Sum Test indicated a statistically significant difference between the two groups with a p-value less than 0.01 (P<<0.01).

Specifically:

  • For normal infants, the ratio of cries with a median complexity below 0.6 to those above 0.6 was approximately 5:1.
  • For deaf infants, this ratio was reversed, with approximately 1:4, indicating a higher proportion of deaf infants exhibiting more complex cry waveforms.

Limitations and Future Work

The authors acknowledge that while their complexity analysis successfully discriminates between normal and deaf infant cries, there is still an "over-lapping portion" between the two domains. This suggests that complexity alone might not be a perfect classifier, and further refinement is needed.

Future work suggested includes:

  • Incorporating More Features: Enhancing the analysis by including other acoustic features such as pitch, formant frequencies, and energy alongside complexity.
  • Reducing Overlap: Developing methods to further reduce the overlap in complexity values between the two groups.
  • Developing an Expert System: The ultimate goal is to create an expert system that can analyze and classify cry modes, potentially leading to diagnostic applications for infant hearing impairment.

Summary updated

BibTeX
@inproceedings{manickam2005complexity,
  author = {Kathiresan Manickam and Haizhou Li},
  title = {{Complexity analysis of normal and deaf infant cry acoustic waves}},
  booktitle = {MAVEBA},
  year = {2005},
}

Loading related content...

Figure