The invention discloses a baby
crying sound translation method based on sound
feature recognition. According to
crying sound characteristic differences of babies in different physiological states,
crying sound characteristic parameters of babies in different physiological states are extracted respectively by using a computer sound
processing technology, wherein the crying sound characteristic parameters are 16 characteristic parameters including 11 characteristic parameters: the tone,
pitch,
loudness, energy, frequency and frequency co-occurrence matrix; on the basis of a BP neural network
algorithm, crying sound sections of the babies in different physiological states are collected, denoising and filtering are carried out on the sound sections, a correspondence relationship between baby crying sound characteristic differences and six kinds of physiological states including hunger, drowsiness, pain, boredom, fear and discomfort is established, and identification results of hunger, drowsiness, pain, boredom, fear and discomfort are provided. In the trained BP neural network, characteristic parameters extracted from any baby crying sound section are inputted and thus an identification result is obtained from the output layer, so that the baby care quality and efficiency are improved and the erroneous determination and delayed determination in baby caring are reduced.