Human infant crying is a crucial survival mechanism, triggering caregiver intervention. While the distress level is clearly communicated, whether the cause of crying (e.g., hunger, discomfort, isolation) is reliably encoded remains debated. Previous studies, often using small and heterogeneous datasets, have yielded conflicting results. Some suggest distinct cry types for different causes, while others propose that cries primarily reflect distress levels. The question of whether cries communicate their cause has significant implications for both caregiver understanding and infant well-being. The research aim was to address this question using a large, well-controlled dataset of longitudinally collected infant cries, investigating not only whether cause is encoded, but also the role of age and individual cry signatures.

The literature surrounding infant cry interpretation is extensive but inconsistent. Early studies suggested that caregivers could distinguish between cry types associated with different causes (e.g., hunger, pain). However, more recent research challenges this, highlighting the difficulty in distinguishing cries based on cause, especially when distress levels are similar. The ability to classify cries has been shown to depend on caregiver experience. Existing cry databases are often small, heterogeneous (mixing healthy and unhealthy infants), or lack longitudinal data, hindering robust analysis. Automated cry analysis using machine learning has also yielded mixed results, with some success in distinguishing pain from other causes but limited success in identifying common causes like hunger or isolation.

This longitudinal study involved 24 full-term, healthy babies (10 girls, 14 boys) recorded at home across four ages: 15 days, 1.5 months, 2.5 months, and 3.5 months. Using omnidirectional microphones, approximately 3600 hours of audio were collected. Parents completed questionnaires for each crying bout, noting the cause, actions taken, and the effective action. Cries were extracted and segmented into 39,201 individual cries. Ten acoustic parameters (including fundamental frequency, various measures of noise and tonal quality, and duration) were measured. Data are publicly available in the EnesBabyCries database. Mixed-effects models and Random Forest classifiers were used to test for relationships between acoustic features and cry cause, age, and individual identity. Two psychoacoustic experiments involved approximately 250 adult listeners (parents and non-parents) who judged the cause of cries in implicit and explicit training paradigms.

Analysis revealed no consistent acoustic differences between cries of boys and girls. However, cries showed systematic changes with age: cries became more tonal (increased harmonicity, decreased entropy) and slightly higher-pitched. A Random Forest classifier could only predict age group with moderate accuracy (~40%). Individual cry signatures were highly identifiable, with a Random Forest classifier achieving ~28% accuracy across all babies and ages (significantly above chance). Individual cry signatures exhibited developmental drift, meaning that though relatively stable, they changed predictably over time. Crucially, analyses found no reliable acoustic differences between cries due to hunger, discomfort, or isolation. Machine learning classifiers performed at chance levels (~36% accuracy) in predicting cry cause, even when controlling for age and individual baby. This was confirmed in psychoacoustic experiments where adult listeners (parents and non-parents) failed to reliably distinguish cry causes, even after implicit or explicit training. There was no evidence for subgroups of babies sharing unique cry cause encoding strategies.

The findings support a graded signal hypothesis for infant cries, where acoustic features primarily reflect distress levels rather than specific causes. The highly discriminable individual cry signatures suggest that familiar caregivers may use additional cues (visual, contextual) beyond acoustic features alone to interpret cry causes. The absence of a universal, discrete cry code for cause might reflect an adaptive flexibility, allowing infants to adapt communication strategies based on both the cause of distress and their caregivers' responses. The observed age-related changes likely reflect maturation of the vocal apparatus and neurocognitive development, potentially including learning to produce cries more effective at eliciting caregiver response.

This study provides strong evidence against the existence of universal, discrete cry types associated with specific causes in healthy infants. Instead, infant cries are characterized by individual acoustic signatures that change predictably with age and likely reflect a graded signal of distress. Future research should explore the role of multi-modal cues in cry interpretation by caregivers and investigate the acoustic features of cries in a wider range of contexts, including those involving pain or illness.

This study focused on short individual cries, potentially overlooking information in the temporal structure or longer cry sequences. The reliance on parental reports to label cry causes is a potential source of bias, though the high correspondence between parental assessments and the actions that stopped crying suggests some validity. The relatively small number of babies limits the generalizability of some findings, and inclusion of more diverse populations and older infants could provide a more comprehensive picture. Finally, this study did not investigate the rarer category of pain cries, so this should be explored in future research.