Skip to main content

Table 2 Available programs used for USV analysis in rodents

From: Beyond the three-chamber test: toward a multimodal and objective assessment of social behavior in rodents

Name

General description and relevant features

References

WAV-file Automated Analysis of Vocalizations Environment Specific (WAAVES)

An automated USV assessment program utilizing MATLAB’s Signal and Image Processing Toolboxes and customizes filters to separate USV calls from noise and assign each USV into one of two categories: 50–55 kHz and 22–28 kHz USVs

Appropriate for rat call analysis

Different test environments (e.g. operant chamber, open field, home cage, etc.) require customized separation criteria

[195]

Automatic mouse ultrasound detector (AMUD)

An algorithm for the automatic detection and extraction of USV syllables which runs on STx acoustic software

The de-noising steps are amplitude-sensitive

Provides information on the detected element frequency, amplitude, and time variables

For detecting USV that are not shorter than 10 ms

[196]

Vocal inventory clustering engine (VoICE)

A classification software that utilizes acoustic similarity relationships between vocal events to generate high dimensional similarity matrices, which are then subjected to hierarchical clustering based on mean frequency and each note's slope, duration, and curvature

Based on pre-defined rules, the syllables are clustered into a limited number [9,10,11,12] of named categories

Includes syntactical similarity quantification to detect changes in syllable patterns across conditions

Independent method is needed to detect and ‘‘clip’’ each syllable into a separate wav file

[190]

Mouse song analyzer (MSA)

A custom MATLAB program based [163] modified from code written by [167] and further developed by [164] for automated, rule-based categorization of syllable shapes

Multi-note syllables are classified based on the number and direction of frequency jumps (or pitch jumps) but not based on the duration, slope, or curvature of each note

The detected syllables are categorized into a limited number [4,5,6,7,8,9,10,11,12,13,14,15] of named categories based on pre-defined rules

Other measured variables include syllable duration, inter-syllable interval, standard deviation of pitch distribution, pitch mean frequency, frequency modulation, and spectral purity [164]

Offers syntax composition and probability analysis to determine the probability of transitioning between different syllable types within a given context, an analysis that enables the identification of repeated syllable patterns (e.g., songs)

[163, 164, 167]

Mouse Ultrasonic Profile ExTraction (MUPET)

An open access MATLAB tool for data-driven analysis of USVs by measuring, learning, and comparing syllable types

MUPET uses an automated and unsupervised algorithmic approach for the detection and clustering of syllable types summarized in the following features:

Syllable detection by isolating and measuring spectro-temporal syllable variables, followed by analyzing overall vocalization features (syllable number, rate and duration, spectral density, and fundamental frequency)

The application of unsupervised machine learning based on k-means clustering to build “syllable repertoire” from the dataset which includes up to several hundreds of the most represented syllable types based on spectral shape similarities within that dataset

Similarity measurement between syllable types of two different repertoires using rank order comparisons in a manner that is frequency-independent

Centroid-based (k-medoids) cluster analysis of syllable types from different syllable repertoires of different datasets to measure the frequency of use of different syllable types across conditions or strains and identify shared and unique shapes

Provides automated time-stamps of syllable events for synchronized analysis with behavior

The option for the user to control features regarding noise reduction, minimum and maximum syllable duration, minimum total and peak syllable energy, and the minimum inter-syllable interval needed to separate rapidly successive notes into distinct syllables

Cannot detect USVs below 30 kHz [197]

[166]

DeepSqueak

A USV detection and analysis software suite based on regional convolutional neural network architecture to detect and categorize USV calls syllables

Packaged with four default detection networks: one general-purpose network, one for mouse USVs, one for short rat USVs and one for long 22 kHz rat USVs

Detecting USVs is done by a region proposal network, which segments the filtered sonogram image into proposed areas of interest with possible USVs, which are then passed to the classification network to determine whether the image contains a call or background noise. The detected USVs are then saved to a detection file along with call variables and classification confidence scores

An option for creating and training custom de-noising secondary networks (by manual annotation of noise vs. call) for identifying noises that might be specific to certain experiments\setups

For syllable clustering, the user can determine which USV features are most important for clustering and adjust three weighted input features that are contour-based: shape, frequency, and duration (thus clustering is amplitude invariant). The number of clusters can be determined by the user using supervised neural networks, or by unsupervised data-based clustering by using k-means on perceptually relevant dimensions of the extracted contour to place calls into a predefined number of clusters

[197]

USVSEG

A program for detecting USV segments (syllables) in continuous sound data containing background noise from several rodent species

Output contains segmented sound files, image files, and spectral peak feature data that can be used for clustering, classification, or behavioral assessment using other toolkits

[198]

VocalMat

A software that uses image-processing and differential geometry approaches to detect USVs in spectrograms, thus eliminating the need for user-defined parameters or costume training of the neural network

VocalMat uses computational vision and machine learning by training a convolutional neural network to classify detected USVs into distinct 11 USV categories or noise

[199]