The Broadband Activity Index

Introduction

The Broadband Activity Index (BBAI) is an acoustic index designed to detect and quantify broadband acoustic activity by identifying ‘clicks’ in audio recordings. Originally developed for monitoring environmental noise such as rain and wind in soundscape recordings, this index has broader applications in environmental acoustic monitoring and bioacoustic research, including quantification of stridulation activity by Orthoptera (crickets, grasshoppers, etc.), non-stridulating pests in stored food or wood, among others. The BBAI follows the principle that most of these sounds have broadband acoustic signatures that appear as vertical “clicks” in a spectrogram. These clicks span multiple frequency bins simultaneously and represent rapid, broadband acoustic events, also known as transient sounds.

Step-by-Step Process

1. Signal Preprocessing

The function begins by preprocessing the audio signal: - DC Offset Removal: Eliminates any constant bias in the signal - High-Pass Filtering: Optional filtering to remove low-frequency noise (default 0 Hz, commonly set to 500 Hz for consistency)

2. Spectrogram Generation

A spectrogram is created using the spectrogram_cutoff() function with: - Frequency Resolution: Determined by the freq.res parameter (default 50 Hz) - Noise Reduction: Row-wise noise reduction is applied - Amplitude Cutoff: Values below the threshold (default -60 dBFS) are removed - dBFS Conversion: Amplitude values are converted to decibels full-scale

3. Click Detection Algorithm

The core detection algorithm works column-wise (time-frame by time-frame) through the spectrogram:

Calculate differences between adjacent frequency bins
Identify contiguous regions where differences are small
Allow small gaps within clicks
Keep only regions longer than minimum click length

Key Detection Parameters:

difference: Maximum amplitude difference (dB) between adjacent frequency bins to be considered part of the same click (default 10 dB)
click.length: Minimum length in frequency bins for a valid click (default 10 bins)
gap.allowance: Maximum gap size (in frequency bins) allowed within a single click (default 2 bins

4. Click Characterization

For each detected click, the algorithm calculates:

Height: Number of frequency bins spanned
Spectral Centroid: Frequency-weighted center of the click
Temporal Position: Time frame location

5. Frequency Band Analysis

The detected clicks are analyzed within four frequency bands, with the following default ranges:

Low-Frequency Activity (LFA): 0-2 kHz
Mid-Frequency Activity (MFA): 2-8 kHz
High-Frequency Activity (HFA): 8-15 kHz
Ultra-Frequency Activity (UFA): 15-22 kHz

6. Index Calculation

The final BBAI value is the percent of spectrogram cells that belong to a click:

BBAI = (Total Click Cells / Total Spectrogram Cells) × 100

Usage Example

if (!require("RthopteraSounds")) install.packages("RthopteraSounds")
library(RthopteraSounds)  
library(Rthoptera)
# Load example data 
data(coryphoda)  
# Basic usage 
result <- broadband_activity2(coryphoda, 
                              freq.res = 100,
                              cutoff = -50,
                              spectrogram = TRUE, 
                              verbose = TRUE)

## Calculating Broadband Activity Index on the left channel... 
## Automatic window length: 960 
## Broadband Activity:  25.7 
## Low-frequency Activity (0 - 2 kHz): 1.1 %
## Mid-frequency Activity ( 2 - 8 kHz): 6.2 %
## High-frequency Activity ( 8 - 15 kHz): 32.1 %
## Ultra-frequency Activity ( 15 - 22 kHz): 42.7 %

## # A tibble: 1 × 17
##   index channel value nclicks prop.clicks click.rate mean.length var.length
##   <chr> <chr>   <dbl>   <int>       <dbl>      <dbl>       <dbl>      <dbl>
## 1 bbai  left     25.7     118         1.3        134        91.9     11589.
## # ℹ 9 more variables: sd.length <dbl>, mean.centroid <dbl>, sd.centroid <dbl>,
## #   var.centroid <dbl>, mean.click.dist <dbl>, lfa <dbl>, mfa <dbl>, hfa <dbl>,
## #   ufa <dbl>

# Advanced usage with custom frequency bands 
result <- broadband_activity2(coryphoda, 
                              freq.res = 100,
                              cutoff = -50,
                              lf_roof = 1.5, # 0-1.5 kHz low freq
                              mf_roof = 6,  # 1.5-6 kHz mid freq   
                              hf_roof = 12, # 6-12 kHz high freq  
                              uf_roof = 20,# 12-20 kHz ultra freq 
                              click.length = 15, # Longer minimum
                              difference = 8,    # Stricter difference threshold       
                              spectrogram = TRUE)

## Calculating Broadband Activity Index on the left channel... 
## Automatic window length: 960 
## Broadband Activity:  24.6 
## Low-frequency Activity (0 - 1.5 kHz): 1.1 %
## Mid-frequency Activity ( 1.5 - 6 kHz): 4.9 %
## High-frequency Activity ( 6 - 12 kHz): 18.4 %
## Ultra-frequency Activity ( 12 - 20 kHz): 40 %

## # A tibble: 1 × 17
##   index channel value nclicks prop.clicks click.rate mean.length var.length
##   <chr> <chr>   <dbl>   <int>       <dbl>      <dbl>       <dbl>      <dbl>
## 1 bbai  left     24.6     103         1.2        117         101     10019.
## # ℹ 9 more variables: sd.length <dbl>, mean.centroid <dbl>, sd.centroid <dbl>,
## #   var.centroid <dbl>, mean.click.dist <dbl>, lfa <dbl>, mfa <dbl>, hfa <dbl>,
## #   ufa <dbl>

# View results table 
print(result$summary)

## # A tibble: 1 × 17
##   index channel value nclicks prop.clicks click.rate mean.length var.length
##   <chr> <chr>   <dbl>   <int>       <dbl>      <dbl>       <dbl>      <dbl>
## 1 bbai  left     24.6     103         1.2        117         101     10019.
## # ℹ 9 more variables: sd.length <dbl>, mean.centroid <dbl>, sd.centroid <dbl>,
## #   var.centroid <dbl>, mean.click.dist <dbl>, lfa <dbl>, mfa <dbl>, hfa <dbl>,
## #   ufa <dbl>

Output Column Explanations

Core Identification Columns

index: The name of the index being calculated. Always “bbai” (Broadband Activity Index). This is useful when merging results with other acoustic indices.

channel: Indicates which audio channel was analyzed (“left”, “right”, “mix”, or “each” for stereo processing).

Primary Activity Measures

value: The main Broadband Activity Index value, expressed as a percentage. This represents the proportion of spectrogram cells that were classified as part of a “click” relative to the total number of cells in the spectrogram. Higher values indicate more broadband acoustic activity.

nclicks: The total number of discrete clicks detected in the recording. A click is defined as a contiguous sequence of frequency bins that meet the detection criteria (minimum length, amplitude difference thresholds, etc.).

Temporal Distribution Measures

prop.clicks: The proportion of time frames that contain at least one click, expressed as a decimal (0-1). This indicates how temporally distributed the clicks are throughout the recording.

click.rate: The rate of click occurrence per second (clicks/second). This provides a temporal density measure of acoustic activity.

mean.click.dist: The mean distance (in time frames) between consecutive clicks. This helps characterize the temporal clustering of acoustic events.

Click Morphology Measures

mean.length: The average length of detected clicks, measured in frequency bins. This indicates the typical frequency span of broadband events.

var.length: The variance in click lengths. High variance suggests clicks of very different frequency spans, while low variance indicates more uniform click sizes.

sd.length: The standard deviation of click lengths. This is the square root of var.length and provides a more interpretable measure of click length variability.

Spectral Centroid Measures

mean.centroid: The average spectral centroid of all detected clicks, expressed in kHz. The spectral centroid represents the “center of mass” of the frequency spectrum for each click, indicating where most of the acoustic energy is concentrated.

sd.centroid: The standard deviation of click centroids. This measures how much the frequency center of clicks varies across the recording.

var.centroid: The variance of click centroids. High values suggest clicks with very different spectral characteristics.

Frequency Band Activity Measures

lfa (Low-Frequency Activity): The percentage of cells classified as clicks in the low-frequency band (0 to lf_roof kHz, default 0-2 kHz). This captures activity in the lower portion of the frequency spectrum, often associated with larger organisms or environmental sounds.

mfa (Mid-Frequency Activity): The percentage of cells classified as clicks in the mid-frequency band (lf_roof to mf_roof kHz, default 2-8 kHz). This band often captures vocal activity from many animals and some mechanical sounds.

hfa (High-Frequency Activity): The percentage of cells classified as clicks in the high-frequency band (mf_roof to hf_roof kHz, default 8-15 kHz). This range often includes insect activity, bird calls, and higher-pitched mechanical sounds.

ufa (Ultra-Frequency Activity): The percentage of cells classified as clicks in the ultra-high frequency band (hf_roof to uf_roof kHz, default 15-22 kHz). This captures very high-frequency sounds like ultrasonic insect calls, bat echolocation, or high-frequency electronic noise. Will be set to NA if hf_roof exceeds the Nyquist frequency.

Stereo Processing Additional Columns

When processing stereo recordings with channel = "each", the function returns separate values for left (_l) and right (_r) channels, plus averaged values (_avg) for each measure:

value_l, value_r, value_avg
n_clicks_l, n_clicks_r, n_clicks_avg
And so on for all measures…

Frequency Band Configuration

Default Thresholds

The frequency bands are defined by the threshold parameters:

Low-Frequency Band (LFA): 0 kHz to lf_roof (default 2 kHz)
Mid-Frequency Band (MFA): lf_roof to mf_roof (default 2-8 kHz)
High-Frequency Band (HFA): mf_roof to hf_roof (default 8-15 kHz)
Ultra-Frequency Band (UFA): hf_roof to uf_roof (default 15-22 kHz)

Automatic Adjustments

The function includes intelligent fallback logic:

uf_roof Adjustment: If uf_roof is set higher than the Nyquist frequency (sampling rate ÷ 2), it is automatically adjusted to the Nyquist frequency to prevent analysis beyond the available frequency range.
hf_roof Adjustment: If hf_roof exceeds the Nyquist frequency, it is adjusted to the Nyquist frequency, and ufa is set to NA since no ultra-frequency band would be available.
Empty Bands: If a frequency band contains no frequency bins (e.g., if thresholds are set inappropriately), the corresponding activity measure will be 0.

Parameter Tuning Guidelines

Detection Sensitivity

Increase click.length: Makes detection more conservative (fewer, longer clicks)
Decrease difference: Makes detection more sensitive (detects subtler amplitude changes)
Adjust cutoff: Lower values (more negative) include quieter sounds

Frequency Resolution Trade-offs

Lower freq.res: Better frequency resolution, longer processing time
Higher freq.res: Faster processing, coarser frequency resolution

Interpretation Guidelines

Activity Level Interpretation

BBAI < 1%: Very low broadband activity
BBAI 1-5%: Low to moderate activity
BBAI 5-15%: High activity
BBAI > 15%: Very high activity (may indicate noise or equipment issues)

Frequency Band Patterns

High LFA: Large organisms, wind noise, equipment noise
High MFA: Crickets
High HFA: Katydids
High UFA: Ultrasonic insects, bats, electronic noise

Morphological Indicators

Large mean.length: Broadband events (impacts, broadband noises)
High var.length: Diverse acoustic events
High mean.centroid: Activity concentrated in higher frequencies
High click.rate: Frequent, brief acoustic events

Limitations and Considerations

Technical Limitations

Sampling Rate Dependency: Higher sampling rates provide access to ultrasonic frequencies
Frequency Resolution Trade-off: Better resolution requires longer processing time
Amplitude Sensitivity: Performance depends on appropriate cutoff thresholds

Biological Considerations

Species Specificity: Different organisms produce different acoustic signatures
Environmental Noise: Background noise can inflate BBAI values
Seasonal Variation: Activity patterns change with environmental conditions

Best Practices

Consistent Parameters: Use identical parameters when comparing recordings
Quality Control: Monitor for equipment-generated artifacts
Validation: Cross-validate with manual inspection of spectrograms
Documentation: Record all parameter settings and environmental conditions

Conclusion

The Broadband Activity Index provides a powerful tool for quantifying acoustic activity across multiple frequency bands. Its ability to detect broadband “clicks” makes it particularly valuable for monitoring non-vocal organisms and environmental disturbances. The enhanced frequency band analysis allows researchers to understand not only the overall level of acoustic activity but also its spectral distribution, providing deeper ecological insights.

By understanding the underlying detection algorithm and properly tuning parameters for specific applications, researchers can effectively use the BBAI for a wide range of acoustic monitoring and analysis tasks.

Francisco Rivas

2025-09-19