Scene-adaptive-based infant cry detection method and system, storage medium and device

Abstract

The present application relates to a kind of based on scene self-adapting baby cry detection method, system, storage medium and equipment, solve the performance deterioration of existing neonatal baby cry detection and recognition method in new environment, poor generalization ability, artificial calibration and training data are insufficient, it is difficult to obtain sufficient detection performance in real clinical environment, often show the technical deficiency of low precision high recall rate, and propose based on scene self-adapting baby cry detection method, system, storage medium and equipment.The method includes the following steps: step S1, based on the scene simulation data enhancement of voice linear addition characteristic: collect voice public data set and collect public baby cry data, utilize public labeled data to carry out data enhancement, and utilize the sample of scene simulation data enhancement to train classifier, learn the cry characteristic under the synthesis of confusion scene;Step S2, scene self-adapting mutual learning;Exponential moving average method and pseudo-label training mode are used, establish a scene self-adapting cry detection learning closed loop, constantly deduce the cry data distribution under new environment, and correct the cry characteristic obtained from step S1, realize environment self-adapting cry detection.

Images