Fourier Transform Based Analysis

This technique relied on the Fourier Transform to determine the frequency sections of the audio signals. This process shared all the parameters of the filter based analysis except the filter order.

The first step of the process was to break the signal into time intervals. The Fourier Transform was then applied to each interval. The resulting frequency data was then broken into a number of section, and low pass filtered to produce an amplitude envelope. This envelope data was averaged and the various frequency section were used to construct a vector of frequency range values. When this process was applied to each interval in the signal, it produced a vector set describing the time and frequency characteristics of the input audio signal.

This process required that all audio signals have the same sampling frequency for the resulting vector sets to be comparable. While birds have been known to vocalize up to 15000 Hz, the vast majority stay within the range of 4000 - 9000 Hz. As such, it was decided to accept a sampling frequency of 22050 Hz to maximize the data set. Since some of the audio recordings obtained were sampled at 44100 Hz, these recordings were converted to 22050 Hz before being processed.