Identification of cracking sound during coffee roasting using neural network

Abstract

Many methods of coffee roasting in the market today are only based on the temperature in the certain time period. However, if the coffee beans have no uniformity in size, weight, and moisture, the roasting process will not produce the consistent results. In this study, the measurement and identification of cracking sounds of coffee beans under roasting are applied to determine the temperature control mechanism. Roaster uses an oven-type controlled by heating element at a temperature of 260°C. In the roasting process, there are the first and second cracking sounds in the time span of 3–10 minutes. Voice Activity Detection is used to identify the cracking sound using Fast Fourier Transform to determine the starting point of sound recording. The data would be learned by the Neural Network to recognize the cracking sounds automatically. The Neural Network can obtain the best result during the period of 1-second recording with success rate of 100%.