Objective 2: Signal Processing Algorithm

Develop signal processing algorithms to automate the analysis of acoustic observations from Habitat Sensor Platforms

Pattern recognition and signal processing algorithms have been developed (Gage and Joo, In Prep.) to interpret acoustic signals collected by acoustic Habitat Sensor Platforms.  Two types of analytical processing have been developed. These analytical domains include 1) a generalized sounds classfication system and 2) automated signature identification (Figure 1).

Figure 1. Two main analysis themes, using data from automated habitat sensor observations.

1) Generalized Sound Classification System

Each sound sample is analyzed to dtermine the spectral energy in one-kHz frequency intervals using the Welch algorithm (Welch 1967). The numberical values resulting from the process provides a quantitative measure of how the acoustic spectral energy is distribtued across the acoustic spectrum (See Figure 2).

Figure 2. The process of quantifying and visualizing acoustic spectral power in one-kHz acoustic frequency intervals

Based on the quantitative analysis of acoustic data, An index of biological activity was developed (Napoletano 2005) to provide a classification of a place relative to its biological composition or its habitat fragmentation based on the amount of energy in different frequency intervals. This has been automated to process and visualize acoustic signals in large digital acoustic libraries.

2) Signature Identification

A central theme of our acoustic monitoring is to be able to identify species based on their acoustic signatures in near real-time. Sound identification using the human ear is the method used to quantify the national breeding bird survey (Ralph et al. 1995). We have developed a method to identify species based on the extraction of signatures of a species from sonogram images (See Figure 3). We match the signature with samples of sound to determine whether the signature occurs in the sample and its frequency of occurrence. The algorithm, developed using MATLAB (Mathworks 2006), matches the image signature file with the image of the acoustic sample. The resultant probability provides evidence of the degree of match between a specific signature and an acoustic sample.

Figure 3. Average 1/2 hourly estimates of spring peeper acoustic signaling for each day in May with 1488 automated recordings of 30 second duration.

<< Previous | | Next >>