Cry analyser to track babies' health

Scientists have developed a new computer-based tool that analyses the cries of babies, searching for clues to potential health or developmental problems.
Subtle acoustic features of a cry, many of them imperceptible to the human ear, can hold important information about a baby's health, researchers said.
A team of researchers from Brown University and Women & Infants Hospital of Rhode Island has developed a new computer-based tool to perform finely tuned acoustic analyses of babies' cries.
The team hopes their baby cry analyser will lead to new ways for researchers and clinicians to use cry in identifying children with neurological problems or developmental disorders.

Slight variations in cries, mostly imperceptible to the human ear, can be a 'window into the brain' that could allow for early intervention, researchers said.
"There are lots of conditions that might manifest in differences in cry acoustics. For instance, babies with birth trauma or brain injury as a result of complications in pregnancy or birth or babies who are extremely premature can have ongoing medical effects," said Stephen Sheinkopf, assistant professor of psychiatry and human behaviour at Brown, who helped develop the new tool.

"Cry analysis can be a noninvasive way to get a measurement of these disruptions in the neurobiological and neurobehavioral systems in very young babies," he said.

The system operates in two phases. During the first phase, the analyser separates recorded cries into 12.5-millisecond frames. Each frame is analysed for several parameters, including frequency characteristics, voicing, and acoustic volume.
The second phase uses data from the first to give a broader view of the cry and reduces the number of parameters to those that are most useful.
The frames are put back together and characterised either as an utterance - a single 'wah' - or silence, the pause between utterances. Longer utterances are separated from shorter ones and the time between utterances is recorded.
Pitch, including the contour of pitch over time, and other variables can then be averaged across each utterance.

In the end, the system evaluates for 80 different parameters, each of which could hold clues about a baby's health.
The paper describing the tool is published in the Journal of Speech, Language and Hearing Research.