Ability evaluation of a voice activity detection algorithm in bioacoustics: A case study on poultry calls

Poultry is one of the most strategic source of human foods. There have been seen some hopeful signs of bioacoustics application to monitor the health condition of this vital food source. One of the obstacles is that the bird's call is combined with some unvoiced sounds and extracting the calls is not easy, especially when the bird is sick. This research is a report on successful application of some of the features involved in extracting healthy and non-healthy birds’ calls from their sound signals. One hundred and twenty birds from two genotypes – Ross and Cobb – were placed in two groups, a control and those challenged with respiratory diseases. They were reared and their sound was recorded daily. The vocal phrases of the recorded audio signals were extracted using the presented algorithm. Results of analysis showed that an increase in age and onset of illness are two factors that cause an error increase. Detection accuracy was calculated at 95% for healthy young birds and 72% for non-healthy birds. A significant part of this error is due to misclassing the calls as non-vocal segments. This meant that 97% of the activities classified as vocal phrases were, in fact, vocal. These results showed that the idea of such an easy-to-implement algorithm could potentially be employed for the coarselevel segmentation of some animal vocalization signals with reliable outputs, which is an essential and primary step in bioacoustics research.