Apple can add to the future AirPods, based on their existing functions. src = “/wp-content/uploads/e82785d68fcda659f741d5fa74cd1bd6.jpg”/> 0 faces> facebookbook
The latest studies of Apple show how the models created for speech recognition can determine the frequency of cardiac contractions by processing by processing by processing by processing by processing by processing phonocardiograms that can be used in Airpods.
on Thursday, a week after the company revealed its study of the humanoid policy, which concerns robotics, Apple published new studies in which health applications were studied. In particular, the assessment of the heart rhythm.
In its latest study entitled “The Foundation Model model, hidden representations for assessing the frequency of cardiac abbreviations for auscultation”, the company's researchers explain how the foundation models can be used to determine the frequency of cardiac contractions from audio recordings, even if they were created for a different purpose.
Apple checked six popular foundation models, including Whisper, which is used to transcript audio to determine whether they can be used outside the alleged purpose. The third -party model CLAP (contrasting language preliminary preliminary) was also tested.
Instead of registering speech, the foundation models were made for processing records of phonocardiogram or heart sounds, for a total of about 20 hours.
the records under consideration belong to the publicly available set of the phonocardiograms of the Circor Digiscope and varyed in the length of 5.1 to 64.5 seconds.
Apple spent tests from about 20 hours of the audio files of the phonocardiogram. Image loan: Apple
The person’s annotator annotated the heart noise in the audio files phonocardiograms, and the files were divided into five second fragments that change every second. The process gave a total of 23,381 fragments of the heart designed to transform into measurement of heart rate.
The company's study explains that its own CLAP model has become better than existing models in this task. “We observed that the ideas from the audio core of the CLAP internal model reached the lowest average absolute error (MAE) in various data gaps, exceeding the basic model trained in standard acoustic features.”
Unlike Whisper, WAV2VEC2 and WAVLM, which the company also checked in the same way, its own CLAP model was trained in audioDades that cover more than simply recorded speech. Various data of training “probably increase its ability to capture non-speech functions related to the heart, contributing to its increase in efficiency,” says the study.
The Apple CLAP model has surpassed other foundation models. Image loan: Apple
Apple researchers say that the foundation models created to process speech “can be effectively adapted for auscultation and evaluating vital signs, offering a reliable and effective alternative to some ordinary methods.” The company's results also show that larger foundation models do not always work better when determining the heart rate.
According to the study, additional accurate setting up of the foundation models can lead to better heart rate estimates. The company's researchers see potential applications in the “pathological analysis of cardiorespirator sounds, which can facilitate the more accurate detection of anomalies, such as arrhythmias and noise.”
as for use in devices, Apple can include this technology in future AirPods models, given the company's interest in health functions. For example, AirPods Pro 2 supports a set of hearing health functions and offers active noise reduction (ANC) through built -in microphones.
Hearing since December 2024 suggested that Apple seeks to realize the sensors of the heart and temperature, among other things. It is possible that the company will use the existing AirPods microphones along with new sensors to more accurately determine the heart rhythm.
