Individualized Own Voice Detection in Hearing Prostheses

The system performs preliminary actions by collecting audio data during a training phase before actual deployment. The algorithm is pre-trained with user-specific voice characteristics through supervised learning, where labeled audio data (marked by the user as own voice or external voice) is used to establish baseline patterns. This preliminary training enables the simplified algorithm to achieve high accuracy without requiring complex real-time processing.

The hearing prosthesis performs self-service by automatically adapting to the user's specific voice characteristics without requiring manual configuration or complex external processing. The device collects audio samples, processes them through the trained algorithm, and continuously refines its own voice detection capabilities based on user feedback and usage patterns, making the system individually optimized without adding external complexity.

The system applies partial action by collecting only the essential audio data needed for training during the initial phase, rather than continuously collecting all possible voice samples. The algorithm processes a sufficient but not excessive amount of training data to achieve the required accuracy threshold, then transitions to efficient real-time detection with minimal additional time investment.

The training data collection and algorithm development is performed as a preliminary action during device setup or initial usage period. Once the individualized model is trained and stored in the hearing prosthesis, the system transitions to rapid real-time voice detection without requiring ongoing extensive data collection, thus minimizing time loss during actual operation.

The system extracts only the most critical time-varying features from the audio signal that are necessary for distinguishing own voice from external voice. Rather than analyzing all possible signal characteristics, the algorithm identifies and processes only the key features (such as fundamental frequency, spectral tilt, and temporal envelope) that provide the highest discriminative power with minimal computational energy expenditure.

The algorithm dynamically adjusts processing parameters based on the input signal characteristics and operating conditions. By changing analysis parameters such as window size, sampling rate, and feature extraction depth according to the current acoustic environment and detection confidence levels, the system optimizes the balance between measurement precision and energy consumption, using more intensive processing only when necessary.