Zoundream's AI-Based Cry Analysis: A Breakthrough in Infant Care and Family Well-being.

Zoundream technology has been built based on over five years of research, in collaboration with pediatricians, clinical researchers, and hospitals around the world. The technology has been used by thousands of families with their babies, for over a milion hours of use, inside families' homes.
The collaboration with some hospitals also led to some scientific papers being published on top tier clinical journals. An extract (and link to full articles) can be found below: These findings will contribute to a better understanding of the interpretation of cry never accomplished until now, emphasizing the clinical potential of cry analysis as an objective non-invasive AI-based tool to enhance the parent-child relationship and ensure family well-being and the proper development of the newborn.

Scientific Journal Publications:

Multi modal Analysis of Infant Cry Types Characterization: Acoustics, Body Language and Brain Signals

This pioneer study examines a multi-modal approach to characterizing infant cry types, utilizing acoustics, body language, and neurophysiological signals like EEG and NIRS. The study collected a realistic dataset from 38 healthy newborns at Hospital Clínic (Barcelona, Spain)under infant’s spontaneous conditions. Machine Learning models, including a specialized Acoustic MultiStage Interpreter (AMSI), were applied to classify cry types.
Results indicated significant differences in cry characteristics across types, with features from acoustics and body language emerging as the most informative for cry classification. The AMSI algorithm achieved a 92% accuracy classifying hunger, sleepiness, fussiness, need to burp and distress.These findings will contribute to a better understanding of the interpretation of cry never accomplished until now, emphasizing the potential of cry analysis to enhance the parent-child relationship and ensure family well-being and the proper development of the newborn.

How can cry acoustics associate newborns’ distress levels with neurophysiological and behavioral signals?

In this innovative research, conducted through a clinical study at Hospital Clínic (Barcelona, Spain), we used for the first time cry audio analysis as a prospective vocal biomarker of newborns' distress states cross-validated with behavioral and brain signals. Data from 25 healthy full-term newborns, including cry acoustics, EEG, NIRS, and video recordings, were analyzed across resting, cry, and distress conditions. The study developed a Deep Learning algorithm that achieved a 93% accuracy in classifying distress levels based on audio features.
Findings indicate that cry features such as pitch and intensity correlate with neurophysiological signals, highlighting infant crying as a potential biomarker of distress quantitatively supporting the assessment of the arousal state of the infant in the future neonatology.

Scientific Conferences:

Quantitative Analysis of Cry Acoustics and Neurophysiological Signals for Assessing the Infant Health Status

This poster presented at the 14th International Newborn Brain Conference in Cork in February 2024 examines the use of cry acoustics and neurophysiological signals as indicators of infant health, aiming to establish cry analysis as a non-invasive biomarker for early neurological assessment.
Data from three groups: full-term, extremely premature, and infants with brain injuries was collected from spontaneous cries and neurophysiological recordings at Hospital Clínic (Barcelona, Spain). Cry acoustic features (e.g., jitter, shimmer, and noise) and brain wave patterns (delta and theta bands) were analyzed, revealing significant differences across the groups. Pathological infants exhibited elevated jitter and shimmer levels, while neurophysiological analyses indicated decreased delta and increased theta band power among infants with brain injuries.

Automatic Detection of Pathological Signs in Newborns Through AI and Cry Analysis for Assessing Infant Health Status

Our latest research presented at the European Academic of Pediatric Societies in Vienna in October 2024 explores the development of an AI-driven tool for the early identification of health conditions in newborns through cry analysis. Traditional assessments of newborn cries, used to detect neurological issues, pain, or discomfort, rely heavily on subjective expertise. This study aimed to create an objective, AI-based algorithm to differentiate between the cries of healthy newborns and those with pathological conditions.
A cry dataset was collected from 58 newborns at Hospital Clínic (Barcelona, Spain), consisting of both healthy and pathological newborns (infection, gastrointestinal, respiratory, etc.). Using a gradient-boosted decision tree model, the algorithm achieved a 94% accuracy in distinguishing between the two groups. Key distinguishing cry features included voice quality, energy, and vocal tract physiological factors.The study demonstrates the potential of cry analysis as a non-invasive, objective tool for early diagnosis, offering significant value in clinical neonatal assessments.