When AI learns to listen: how researchers are decoding baby cries to help new parents

Imagine if a smartphone could listen to a newborn’s cry and indicate whether the baby needs feeding, sleep, or comfort, removing the guesswork for exhausted parents. 

This is not science fiction. It is the vision behind LetBabyTalk, an AI-powered multilingual parenting app trained on more than 1,000 baby cries from infants aged zero to 12 months. Developed by researchers connected to Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), the project applies causal AI research to real-world parenting challenges. 

At the center of the work is Yuewen Sun, a postdoctoral associate at MBZUAI and chief executive officer of Cradle AI, the startup behind LetBabyTalk.  

By recording a baby’s cries, LetBabyTalk identifies likely needs such as hunger, sleepiness, or discomfort and provides personalized advice based on childcare experience, making parenting easier and more intuitive. 

Human-centered AI: improving lives and society safely 

Cradle AI’s name reflects its mission: to act as a “cradle” for nurturing humans through trustworthy AI. The company focuses on underserved everyday challenges such as family care and education. 

“We saw a gap between advanced AI research and creating solutions based on the real needs of people,” Sun says. “Cradle AI exists to bridge that gap.” 

The idea for LetBabyTalk emerged from conversations with new parents who described exhaustion, anxiety, and stress when they could not understand why their baby was crying. 

“Many parents told us that even small uncertainties accumulate into significant stress,” Sun explains. “We felt that if AI could help parents better understand their babies, even in a small way, it could meaningfully improve their daily lives.” 

Those insights led the team to focus on a clear, human problem where AI could offer immediate value – fulfilling Cradle AI’s broader commitment to “AI for good,” by moving research beyond academic papers and into daily life.

How the technology works 

LetBabyTalk uses supervised machine learning to analyze baby cries and identify patterns associated with common infant needs. The system processes audio through a pretrained foundation model, which the team fine-tuned using curated datasets and labels provided by experienced parents.  

The system is trained on approximately 1,000 baby cry samples and, unlike other similar models, does not rely on audio data alone. 

The team interviewed parents and educators, who said that even a small reduction in uncertainty could make a meaningful difference during the early months of parenting, particularly for first-time parents. 

Several AI-powered “baby translator” and “cry analyzer” apps claim to interpret infant cries, but Cradle AI found many to be less reliable due to a reliance on AI algorithmic analysis alone.  

“We evaluate LetBabyTalk using both public benchmarks and our own real-world data,” Sun explains. “Specifically, we test performance on publicly available infant-cry datasets such as the Baby Chillanto database by using the same audio inputs to compare accuracy across different apps when comparable input formats are supported.” 

From causal AI research to real-world parenting challenges 

Sun works closely with Kun Zhang, MBZUAI’s associate department chair of machine learning research and visiting professor of machine learning, who co-founded the startup and serves as board chair. Zhang’s research focuses on causal AI: a branch of AI that goes beyond pattern recognition to understand cause-and-effect relationships.  

“My goal is to build AI systems that are interpretable, reliable, and controllable in real deployment settings,” Sun says. “These systems can meaningfully support decision-making and have a positive impact on society.” 

With a Ph.D. in control science and engineering, Sun brings a background in systems theory, optimization, and data-driven modeling. That foundation has shaped her focus on AI systems designed to function responsibly in real-world environments such as healthcare, education, and early childhood support. 

The team and the startup journey 

Founded in 2024, Cradle AI received support and mentorship from MBZUAI’s Incubation and Entrepreneurship Center. 

“Here, cutting-edge research and real-world innovation connect closely,” Sun says. “That makes it possible to move quickly from ideas to impact, while staying grounded in safe and ethical AI.” 

Sun serves as CEO, overseeing strategy, product direction and research-to-product translation while remaining actively involved in technical development. 

The team includes Yu Yao, a lecturer at the University of Sydney, as scientific adviser; MBZUAI research assistant Yukang Wong, as chief operating officer; and Minghao Fu, a Ph.D. student at the University of California, San Diego, as chief technical officer. 

Beyond baby care: education and personalized AI 

In five years, Sun hopes Cradle AI will be known for translating strong AI research into products people trust and use, expanding from infant care into education. 

“One of the biggest problems in education is that most systems target the ‘average’ student,” she says. “Real students differ significantly in ability, pace, and needs.” 

By modeling individual developmental trajectories, the research aims to identify when personalized interventions can have the greatest impact; moving education away from one-size-fits-all solutions.

Looking ahead: AI parents can trust 

The team plans to refine LetBabyTalk through user feedback, expand data coverage ethically, and ensure reliable performance across diverse family contexts. 

As AI moves from research labs into daily life, tools like LetBabyTalk show how technology can support parents during one of life’s most demanding stages. That impact may begin with something as simple, and as powerful, as understanding a baby’s cry. 

 LetBabyTalk is available for free on iOS and Android.