Imagine capturing a busking performance, only to have car horns overpower the music — or filming a vlog in a café where ambient noise makes conversations nearly impossible to hear.
Audio Eraser, introduced in the Galaxy S25 series, was designed to address such situations by reducing unwanted background noise and enhancing desired sounds to help users create more immersive videos.1 Galaxy AI features like Audio Eraser — developed with a deep understanding of user needs — was made possible thanks to the proactive efforts of Samsung Research and the Mobile eXperience (MX) Business at Samsung Electronics.
Samsung Newsroom sat down with audio technology experts from the AI Solution Team at Samsung Research to learn more about the development of this innovative feature.
▲ (From left) Kyoungbo Min, Hejung Yang, Hosang Sung and Jiwon Kim from the AI Solution Team at Samsung Research
Source Detection: The Future of Next-Generation Audio Technology
Audio Eraser enables users to adjust sounds within a video by removing unwanted noise and enhancing the audio they want to highlight. The feature quickly scans a video to detect and separate sounds by type, and can be used not only on clips recorded by the user but also on videos that have been shared or received.
Just as Object Eraser improved image editing and usability with its debut in the Galaxy S21 series, Audio Eraser was developed to deliver a more advanced and seamless multimedia experience across both audio and video.
▲ Audio Eraser allows users to adjust sounds by type within a video.
Developing Sound Source Detection and Separation Models
Audio Eraser recognizes six types of sound — voices, music, wind, nature sounds, crowd noise and ambient noise. Using sound source detection technology, the feature quickly identifies what sounds are present and where they occur in a video. Then, through sound source separation technology, it isolates and classifies each sound into distinct categories.
To build AI models capable of detecting and isolating these audio types, the research team began by collecting a robust set of training data. They simulated a wide range of real-world video recording scenarios to generate diverse audio datasets — and even went into the field to capture everyday sounds firsthand.
“Wind was particularly challenging to work with,” said Hejung Yang from the AI Solution Team at Samsung Research. “Besides refining our wind simulation technology, we enhanced our dataset by recording actual wind sounds outdoors whenever strong winds occurred — even after work or on weekends.”
In addition, the team devoted extensive time to improving the performance of the sound source separation model by carefully reviewing countless video clips multiple times.
“Each developer compared and analyzed more than 1,000 audio samples every week under a variety of conditions,” said Jiwon Kim. “Through continuous experimentation, we worked to identify the most effective sound source separation model to ensure consistent, high-quality results.”
▲ Samsung researchers work on developing audio technology for Audio Eraser.
Delivering a New User-Centric Listening Experience
As Audio Eraser runs on-device, it offers an enhanced user experience with real-time editing and improved privacy protection. Samsung’s long-standing expertise in developing competitive on-device AI capabilities across various features played a key role in Audio Eraser’s success.
“We focused on developing AI models and algorithms capable of fast, on-device processing,” said Hosang Sung. “A great deal of effort went into creating an optimized AI solution that runs smoothly with low power consumption.”
Supported by a strong AI research infrastructure, Samsung Research continues to strengthen its capabilities in advanced audio technologies.
“We are committed to developing next-generation solutions that elevate the usability of mobile devices,” said Hoonyoung Cho, Vice President and Head of the AI Solution Team at Samsung Research. “Through sound control and audio enhancement technologies, we aim to deliver a truly selective and personalized listening experience.”
Integrated Collaboration With Product Development Teams
Turning advanced research into a consumer-ready solution brought a new set of challenges. For instance, when editing video and audio simultaneously, playback needed to remain smooth and uninterrupted. Additionally, processing videos longer than an hour required specialized technology capable of completing the task within the same timeframe.
To address these challenges, Samsung Research and the MX Business worked together — from proposing user-centric ideas to repeatedly testing software optimization techniques and evaluating audio quality. Together, they explored multiple approaches to determine the most effective solution.
The cross-functional collaboration had already proven successful in the development of the Galaxy Buds3 series (Buds3, Buds3 Pro). As the needs for listening in noisy environments grows — for everything from music and media experiences to listening to clearer phone calls — Active Noise Cancellation (ANC) has become an increasingly essential feature in wireless earbuds. Samsung raised the bar by developing its own Adaptive ANC technology. The Galaxy Buds3 series adapts to the fitting conditions of different users and changes in fit caused by prolonged wear to deliver optimized ANC performance.
Since earbuds are worn directly in the ear, close coordination between hardware and software teams was essential to finalize product specifications and fine-tune the ANC technology.
“We developed multiple algorithms in advance to accommodate different hardware conditions,” said Kyoungbo Min. “Through extensive simulations, we prepared a solution that could be quickly adapted to any chipset or device. The development process involved iterative revisions and improvements, but the close collaboration between advanced research and product teams helped accelerate commercialization.”
Sound source separation is a core area of next-generation audio technology, and Samsung continues to advance its foundational innovations in this field. With Audio Eraser making full use of sound source separation, the company is preparing to expand into a range of new applications. The ongoing evolution of Galaxy AI is expected to deliver even more practical and intuitive features for consumers.
1 Results may vary per video depending on how sounds present in the video. Samsung Account login required. Certain types of sound can be detected such as voices, music, wind, nature, crowd and noise. The actual sound detection may vary depending on audio source and the condition of the video. Accuracy of results is not guaranteed.
