RESUMO
Auditory prostheses provide an opportunity for rehabilitation of hearing-impaired patients. Speech intelligibility can be used to estimate the extent to which the auditory prosthesis improves the user's speech comprehension. Although behavior-based speech intelligibility is the gold standard, precise evaluation is limited due to its subjectiveness. Here, we used a convolutional neural network to predict speech intelligibility from electroencephalography (EEG). Sixty-four-channel EEGs were recorded from 87 adult participants with normal hearing. Sentences spectrally degraded by a 2-, 3-, 4-, 5-, and 8-channel vocoder were used to set relatively low speech intelligibility conditions. A Korean sentence recognition test was used. The speech intelligibility scores were divided into 41 discrete levels ranging from 0 to 100%, with a step of 2.5%. Three scores, namely 30.0, 37.5, and 40.0%, were not collected. The speech features, i.e., the speech temporal envelope (ENV) and phoneme (PH) onset, were used to extract continuous-speech EEGs for speech intelligibility prediction. The deep learning model was trained by a dataset of event-related potentials (ERP), correlation coefficients between the ERPs and ENVs, between the ERPs and PH onset, or between ERPs and the product of the multiplication of PH and ENV (PHENV). The speech intelligibility prediction accuracies were 97.33% (ERP), 99.42% (ENV), 99.55% (PH), and 99.91% (PHENV). The models were interpreted using the occlusion sensitivity approach. While the ENV models' informative electrodes were located in the occipital area, the informative electrodes of the phoneme models, i.e., PH and PHENV, were based on the occlusion sensitivity map located in the language processing area. Of the models tested, the PHENV model obtained the best speech intelligibility prediction accuracy. This model may promote clinical prediction of speech intelligibility with a comfort speech intelligibility test.
RESUMO
Without expert coaching, inexperienced exercisers performing core exercises, such as squats, are subject to an increased risk of spinal or knee injuries. Although it is theoretically possible to measure the kinematics of body segments and classify exercise forms with wearable sensors and algorithms, the current implementations are not sufficiently accurate. In this study, the squat posture classification performance of deep learning was compared to that of conventional machine learning. Additionally, the location for the optimal placement of sensors was determined. Accelerometer and gyroscope data were collected from 39 healthy participants using five inertial measurement units (IMUs) attached to the left thigh, right thigh, left calf, right calf, and lumbar region. Each participant performed six repetitions of an acceptable squat and five incorrect forms of squats that are typically observed in inexperienced exercisers. The accuracies of squat posture classification obtained using conventional machine learning and deep learning were compared. Each result was obtained using one IMU or a combination of two or five IMUs. When employing five IMUs, the accuracy of squat posture classification using conventional machine learning was 75.4%, whereas the accuracy using deep learning was 91.7%. When employing two IMUs, the highest accuracy (88.7%) was obtained using deep learning for a combination of IMUs on the right thigh and right calf. The single IMU yielded the best results on the right thigh, with an accuracy of 58.7% for conventional machine learning and 80.9% for deep learning. Overall, the results obtained using deep learning were superior to those obtained using conventional machine learning for both single and multiple IMUs. With regard to the convenience of use in self-fitness, the most feasible strategy was to utilize a single IMU on the right thigh.