Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Identification of Background Noise by Mel-scale Frequency Cepstral Coefficients (MFCC) and Vector Quantisation (VQ) for Piglets Sound Detection
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2018 ASABE Annual International Meeting 1800780.(doi:10.13031/aim.201800780)
Authors: Jheng-Hong Hsieh, Ching-Lu Hsieh
Keywords: Sound identification, Piglet sound, Mel-scale Frequency Cepstral Coefficients (MFCC), Vector Quantisation (VQ), LBG algorithm.
Abstract. Swine is a very important animal husbandry in Taiwan. During swine growth, piglet is a crucial stage that will significantly influence its market value. Meanwhile, animal welfare is also an important factor which eventually affects the efficiency of animal growth and meat quality. This study applied a sound recorder with temperature and humidity logger to set up a noninvasive monitor system for piglet stress detection during pigpen change or weaning stage. The monitor system was set up in lab field (a pigpen) after calibration. A 24-hr record showed background information in sound, temperature, and humidity of the pigpen. After three batches of experiments, 270 hr sounds were recorded for nine piglets. From the sound clips, six sound clusters were picked out manually for further analysis. They were piglets, human speech, birds, factory operation, motorcycle pass, and silent sound. Each sound group contained 50 files about 10 second duration. Files were categorized into training set and validation set of half of the numbers for each cluster. Sound clips were extracted features with MFCC which contained framing, windowing, DFT, Mel frequency warping, Log transform, and IDFT. After feature extraction, they were matching with VQ and the shortest Euclidean distance to each cluster was assigned to its sound cluster for recognition. Results showed that recognition accuracy for two clusters in piglet VS human speech of 78%, piglets VS silent sound of 91%, piglets VS factory sound of 85%, piglets VS bird sound of 88%, and piglets VS motorcycle sound 89%. When piglets sounds were compared with those 250 background sound files, the recognition accuracy were 92%.
(Download PDF) (Export to EndNotes)