NADE dataset for figures

A .csv containing data used to produce the figures in the NADE paper.

This dataset contains information about mixed audio samples created by combining bird vocalizations (foreground) with background environmental recordings. Each row represents one synthesized audio clip along with associated metadata and acoustic measurements. Note that amplitudes are expressed in decibels (dB), relative to full scale (dBFS).

Column Name
Description

bg_filename
The filename of the background audio clip used in the mixture.

fg_filename
The filename of the foreground vocalization audio file added to the background.

species_name
Common name of the species corresponding to the foreground vocalization.

bg_datetime
Date and time of the background recording, typically derived from the file metadata.

output_filename
Name of the resulting mixed audio file combining the background and foreground clips.

fg_multiplier
Linear gain multiplier applied to the foreground vocalization before mixing. This value directly controls the foreground amplitude.

scaled_fg_amplitude (dB)
The amplitude of the scaled (post-multiplier) foreground vocalization in decibels (dB).

fg_confidence
Model confidence score (0–1) for detecting the unaltered foreground vocalization before adding background noise.

confidence
Model confidence score (0–1) for detecting the mixed (foreground + background) audio.

location
The site identifier of the background file.

bandpass_dB
Amplitude (in dB) of the background noise measured within the species-specific vocalization frequency band. Represents how noisy the background is in the relevant frequency range.