A central theme of our acoustic monitoring is having the ability to identify species based on their acoustic signatures in near real-time. A sensor data stream is a time series comprising continuous or periodic sensor readings. Typically, readings taken from a specific sensor can be identified, and each reading appears in the time series in the order acquired. These sequences can be clustered and fused with other data to support species detection and classification. Classification attempts to accurately recognize which species produced a particular vocalization, while detection indicates the likelihood that an acoustic clip contains a song voiced by a particular species.

Figures 1a and 1b depict two common methods for visualizing an acoustic clip. Figure 1a shows an oscillogram that depicts a signal's normalized amplitude over time. Figure 1b shows the same clip plotted as an acoustic spectrogram. A spectrogram depicts frequency on the vertical axis and time on the horizontal axis. Color indicates the intensity of the signal at a particular frequency and time, where frequency is measured in cycles per second or hertz (Hz). Spectrograms are useful for visualizing acoustic signals in the frequency domain.

Plotted in Figures 1c and 1d is the power spectral density (PSD) for the signal. Power spectral density depicts the signal strength, or power, found in different frequency bands. Figure 1c shows the PSD at a fine granularity while Figure 1d plots the PSD integrated over bins with a 200Hz width. As shown, the spring peeper (Pseudacris crucifer crucifer) signals at approximately 3 kilohertz (kHz).

Visual and other representations of acoustic and other sensor signals can be used to enable automated classification and detection of acoustic events, including classification and detection of vocalizing species. Moreover, such signal processing and analysis enables the computation of numerous indices and fusing of different sensor data types that can be used for ecosystem assessment.