Abstract
Mouse communicative behaviors are referred to as ultrasonic vocalizations (USVs). USVs are relevant to human health and are commonly recorded with the Avisoft-SASLab Pro recording system and the DeepSqueak system however, no study has compared DeepSqueak’s accuracy to Avisoft-SASLab Pro’s so it is unclear which is the optimal program. We used male and female C57BL/6, FVB.129, and FVB mice and assessed the total USVs detected, USV duration, fundamental frequency, and within system consistency for each system. We found that for all strains assessed, there were extremely high correlations for the quantity of vocalizations that were detected between Avisoft-SASLab Pro and DeepSqueak. Furthermore, no significant differences in USV detection between systems were found for any of the strains. A partial correlation was run to control for the strain, sex, and age covariates and found that no covariate substantially influenced the relationship between the USVs detected in Avisoft-SASLab Pro and those detected in DeepSqueak. We also investigated the duration and fundamental frequency of vocalizations between systems and found medium to large correlations across all strains and several significant differences. We next found that DeepSqueak has higher internal consistency than Avisoft-SASLab Pro when assessing USV duration and fundamental frequency. Lastly, we assessed the average time taken to analyze a file and found DeepSqueak to be the faster system. Altogether, these findings demonstrate that while both DeepSqueak and Avisoft-SASLab Pro are accurate and reliable, DeepSqueak is a more efficient USV analysis program.
Keywords: Autism spectrum disorder, Methods, Epilepsy, Communication, Reproducibility
1. Introduction
Communicative behaviors in mice are known as ultrasonic vocalizations (USVs). Mouse models of neurodevelopmental conditions that present with communication deficits, such as Autism Spectrum Disorder (ASD), display alterations in vocalizations similar to what is seen in humans [1, 2]. Therefore, USVs are a valuable tool in developmental research. Numerous systems have been created to analyze vocalizations. However, these systems are rarely compared to one another, so it is unclear which is the optimal (most accurate and efficient) system to use. Furthermore, when recording programs have been compared, significant differences in accuracy and reliability between systems have been found [3–6]. Therefore, USV analysis programs cannot necessarily be used interchangeably, nor their results interpreted uniformly, as results analyzed in one system may not be reproduced in another. Thus, there is a strong need for direct comparisons of USV analysis programs to identify optimal systems that are accurate, consistent, and efficient.
Avisoft-SASLab Pro and DeepSqueak are two prominent USV analysis programs. Avisoft-SASLab Pro was developed in the early 90s and quickly became a gold standard in the vocalization field due to its accuracy, reliability, detection range, and relative ease of use. Due to this, it has been extensively and routinely used in a myriad of publications [2, 7, 8]. Conversely, the DeepSqueak system was developed much more recently (in 2019) but has nonetheless quickly become assimilated in the field [6]. Both vocalization systems have numerous similarities as they utilize semi-automated algorithms to expediate USV analysis and are commonly used. Furthermore, when Avisoft-SASLab Pro and DeepSqueak have been compared to other recording systems such as the Mouse Song Analyzer, Ultravox, or MUPET, they have routinely outperformed their competition, boasting unmatched accuracy [3–6]. However, DeepSqueak and Avisoft-SASLab Pro have never been directly compared to one another, so the respective benefits, drawbacks, relative accuracy, and reliability of two of the most popular vocalization analysis systems are unknown. In the present study, we directly compared DeepSqueak to Avisoft-SASLab Pro and assessed the total quantity of vocalizations detected, USV duration and fundamental frequency, as well as within system consistency in order to determine which system is optimal.
2. Materials and methods
2.1. Animals and housing
USVs from C57BL/6, FVB.129, and FVB neonates were used in the present study. The sample sizes and day of USV analysis are listed in Table 1, altogether forming a total n of 80 which is in accordance with the apriori power analysis we performed (G*power). Animals were housed in Baylor University’s animal vivarium under standard laboratory conditions (22 °C room, 12 h light/dark cycle, and ad libitum food and water access). All procedures were in accordance with Baylor University’s Institutional Animal Care and Use Committee and the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals.
Table 1.
Mean analysis time and within system reliability
| Sample Sizes | |||||
|---|---|---|---|---|---|
| C57 BL/6 – PD 12 | FVB.129 – PD 11 | FVB PD 8 | |||
|
| |||||
| Males | Females | Males | Females | Males | Females |
| 19 | 17 | 15 | 9 | 11 | 9 |
2.2. Ultrasonic vocalizations
An isolation-induced ultrasonic vocalization paradigm was used to elicit USVs from neonates [2, 9]. Prior to USV recording, pups were separated from their parents and placed into a new housing cage with fresh bedding that was warmed to ambient nesting temperature. The pups were then individually removed from the holding cage and placed into a clean cage within a sound-attenuated chamber. USVs were recorded for 2 min using a broad-spectrum condenser microphone (CM16/CMPA, Avisoft Bioacoustics, Glienicke, Germany) and a recording interface (UltraSoundGate 116Hb, Avisoft Bioacoustics). When the testing period concluded, the pup was removed from the test chamber and returned to its home cage.
2.3. Avisoft-SASLab pro analysis
The Avisoft-SASLab Pro software (Avisoft SASLabPro, RRID: SCR_014438) was used to transform the .wav files into spectrograms. The specific parameters are as follows and were kept consistent across all USV files: a fast Fourier transformation (FFT) length of 1024, time window overlap of 75%, a 100% hamming window, a time resolution of 1 ms, and a sampling frequency of 250 kHz. Furthermore, the automatic parameter measurement feature was enabled, reducing the USV analysis time. These parameters were selected to be in accordance with established literature [2]. Once spectrograms were generated, each file was then cleaned by a trained experimenter which consisted of identifying USVs using visual and auditory cues and trimming the detection bars around the USV to ensure accurate spectral and temporal characteristic measurements, as well as removing any background noise.
2.4. DeepSqueak analysis
DeepSqueak was downloaded from Github (https://github.com/DrCoffey/DeepSqueak) and accessed via Matlab. The USV .wav files were imported into the DeepSqueak program and the analysis parameters were set as follows: a total analysis length of 0, an analysis chunk length of 6, a frame overlap of .0001 s, a frequency low cut off of 30 kHz, a frequency high cut off of 120 kHz, and a score threshold of 0, with detection being set on “high recall.” These parameters were set to be in accordance with the established literature [6]. Once the spectrogram was generated, the files were manually processed, whereby a trained experimenter individually inspected each detected noise in the .wav file then accepted or rejected it based off of visual and auditory cues. If a detected vocalization was not fully contained by the automatic detection boxes, then the detection boundaries would be redrawn in order to encompass the entirety of the USV.
2.5. System consistency analysis
Both the Avisoft-SASLab Pro and the DeepSqueak analysis systems render automatic detection boundaries around each vocalization. While many of these detection boundaries are accurate, some do need to be adjusted to ensure accurate temporal and spectral assessments. For Avisoft-SASLab Pro, this consists of trimming detection bars “error bars” around the USV whereas DeepSqueak requires the detection boundaries to be redrawn by the user. Manually rendered boundaries around a call may introduce slight variability in the assessment of the duration and fundamental frequency which in turn may lead to problems with internal consistency. To test this, we calculated Avisoft-SASLab Pro’s and DeepSqueak’s internal consistency by assessing then reassessing a file in each system. The total quantity of calls produced, duration, and fundamental frequency were compared across the first assessment and the second assessment. The average length of time required to analyze a file in each system was also computed.
2.6. Statistical analysis
IBM SPSS Statistics 21.0 (IBM, USA) and GraphPad Prism 7 software (La Jolla, CA) were used for statistical analysis. To determine the accuracy of DeepSqueak compared to Avisoft-SASLab Pro, correlations for each strain assessed (C57BL/6, FVB.129, FVB) were performed. To assess whether Avisoft-SASLab Pro or DeepSqueak detected a significantly different quantity of USVs from one another, independent t-tests or Mann Whitney U tests per each strain were run. A partial correlation analysis was then conducted to control for the sex (male, female), strain (C57BL/6, FVB.129, FVB), and age (PD 8,11,12) covariates. A zero-order correlation (a correlation that did not control for covariates) was also calculated. This was done in order to determine the influence each covariate had on the relationship in USV detection between Avisoft-SASLab Pro and DeepSqueak. We also assessed the USV fundamental frequency and duration in each system (specifically comparing the mean parameter of entire element for the fundamental frequency in Avisoft-SASLab Pro with the fundamental frequency generated in DeepSqueak, due to their similarity in measurement) applying the same statistical approach detailed above. T-tests and correlations were preformed to assess within file consistency as well as the average duration required to assess a file in both systems. A value of p < .05 was considered significant for each statistical test and the figures depict the mean ± standard error of the mean (SEM). The effect sizes for the correlations are as follows: small effect r = 0.1, medium effect r = 0.3, large effect r = 0.5.
3. Results
3.1. USVs detected in Avisoft-SASLab Pro and DeepSqueak
Large correlations were found for the quantity of USVs detected in Avisoft-SASLab Pro and those detected in DeepSqueak for C57, FVB.129, and FVB mice (r(70) =.97, p<.001, r(46) =.99, p<.001, r(38) =.90, p<.001 respectively) (Fig. 1 A,C,E). Similarly, no significant differences were found between systems in the C57, FVB, and FVB.129 strains (t(70) =.13, p=.90, t(46) =.69, p =.49, t(38) =.33, p =.75) (Fig. 1 B,D,F). To determine the strength of the relationship between the number of USVs detected by the two systems whilst controlling for sex (male, female), strain (C57BL/6, FVB.129, FVB) (Fig. 1G) and age of the animal (PD 8,11,12) a partial correlation was run. There was a large positive correlation between the number of USVs detected across the two systems, which was significant (r (75) = 0.95, p < 0.001). The zero-order correlation also showed that there was a large, significant correlation between the number of USVs detected across the two systems (r (75) = 0.96, p < 0.001), indicating that the covariates had little influence in controlling for the relationship between the two systems (Fig. 1G).
Fig. 1. USVs detected in Avisoft-SASLab Pro and DeepSqueak.
(A,C,E). There are large correlations between Avisoft-SASLab Pro and DeepSqueak for the total quantity of vocalizations detected for (A) C57BL/6, (C) FVB.129, and (E) FVB mice. (B,D,F) When the total quantity of vocalizations emitted per each strain were assessed, no statistically significant differences between systems were found for (B) C57BL/6, (D) FVB.129, and (F) FVB mice. (G). Upon controlling for the variables of sex, strain, and age, a large correlation of r =.95 was found. (G). The zero-order correlation of .96 indicates that none of these covariates substantially influences the relationship between USVs detected in Avisoft-SASLab Pro and those detected in DeepSqueak. The data points represent the mean and the error bars represent the standard error of the mean.
3.2. USV duration in Avisoft-SASLab Pro and DeepSqueak
When assessing USV duration, we found large correlations between systems for C57, FVB.129, and FVB mice (r(70) =.88, p<.001, r(46) =.69, p<.001, r(38) =.72, p<.001) (Fig. 2 A,C,E). However, DeepSqueak detected longer USVs than Avisoft-SASLab Pro for C57, FVB.129, and FVB mice (t(70) =6.65, p<.001, t(46) =4.14, p<.001, U= 48, p<.001). (Fig. 2 B,D,F). There was a large positive correlation between the average duration of USVs detected across the two systems, which was significant (r (75) =0.59, p <.001). However, the zero-order correlation was significantly larger (r (75) =0.91, p <.001), with age accounting for .79 of the variance. Therefore, while the strain and sex covariates have little influence on the average duration of the USVs detected between systems, the age of the mouse does influence the relationship between Avisoft-SASLab Pro and DeepSqueak the for the duration of the vocalizations (Fig. 2G).
Fig. 2. USV duration in Avisoft-SASLab Pro and DeepSqueak.
(A,C,E). Large correlations between systems were found for (A) C57BL/6, (C) FVB.129 and (E) 129 mice. (B,D,F). However, DeepSqueak detected significantly longer USVs than Avisoft-SASLab Pro per each strain. (G). Upon controlling for the variables of sex, strain, and age, a large correlation of r =.59 was found. (G) The zero-order correlation of .91 revealed that age substantially influences the relationship between USVs detected in Avisoft-SASLab Pro and those detected in DeepSqueak, whereas the sex, and strain variables did not substantially influence this relationship. The data points represent the mean and the error bars represent the standard error of the mean. * = p < .05, ** = p < .01, *** = p < .001.
3.3. USV fundamental frequency in Avisoft-SASLab Pro and DeepSqueak
For fundamental frequency, we found a medium correlation between systems for the C57BL/6 strain and a large correlation for FVB.129 and FVB mice (r(70) =.38, p=.02, r(46) =.94, p<.001, r(38) =.87, p<.001) (Fig. 3 A,C,D). We also found that DeepSqueak detected a higher fundamental frequency than Avisoft-SASLab Pro for C57 mice, with no differences present between systems for FVB.129 and FVB mice (U=51, p<.001, t(46) =1.78, p=.08, U=99, p<.01) (Fig. 3 B,D,F). When assessing the fundamental frequency of USVs, we found a medium positive correlation between the average fundamental frequency of USVs detected across the two systems, which was significant (r (75) = 0.38, p < 0.001). However, the zero-order correlation was smaller (r (72) = 0.01, p < 0.001), with age accounting for .43 of the variance and strain accounting for .40 of the variance. Therefore, while the sex covariate had little influence on the average fundamental frequency of the USVs detected between systems, the age and strain variables did substantially influence the relationship between Avisoft-SASLab Pro and DeepSqueak for the fundamental frequency of vocalizations (Fig. 3G).
Fig. 3. USV fundamental frequency detected in Avisoft-SASLab Pro and DeepSqueak.
(A). There is a medium correlation between systems for C57BL/6 mice. (B). DeepSqueak detected a significantly higher fundamental frequency of USV than Avisoft-SASLab Pro for C57BL/6 mice. (C). There is a large correlation between systems for FVB.129 mice. (D). There is no statistically significant difference in the quantity of USVs detected between Avisoft-SASLab Pro and DeepSqueak for FVB mice. (E). There is a large correlation between systems for the FVB mice. (F) There is no statistically significant difference in the quantity of calls detected between Avisoft-SASLab Pro and DeepSqueak for FVB mice. (G). Upon controlling for the variables of sex, strain, and age, a medium correlation of r =.38 was found. (G) The zero-order correlation of .01 revealed that the strain and age of mice substantially influenced the relationship between USVs detected in Avisoft-SASLab Pro and those detected in DeepSqueak, whereas the sex variable did not substantially influence this relationship. The data points represent the mean and the error bars represent the standard error of the mean. * = p<.05, ** = p<.01, *** = p<.001.
3.4. System consistency and analysis time
We first assessed the average time required to analyze files in each system. We split a subset of files into two categories: “low vocalizer” with <200 USVs and “high vocalizer” with >200 USVs. We found that DeepSqueak took significantly less time than Avisoft-SASLab Pro to analyze both high and low USV files (t (38) = 13, p < .0001, t (38) = 8.376, p < .0001). (Fig. 4A,B). When assessing the within file consistency for USV detection we found that both Avisoft-SASLab Pro and Deepsqueak detected the same amount of calls on the first assessment and on the second assessment (Fig. 4C).
Fig. 4. Mean analysis time and within system reliability.
(A,B). DeepSqueak analyzes files with both low and high USVs faster than Avisoft-SASLab Pro. (C). When assessing within system reliability for USV count data there were no differences between the quantity of USVs detected in the first and second assessment for either program. (D-F). For call duration, there was no significant differences between the first and second assessments and large correlations were found between assessments in each system, however, DeepSqueak had higher overall correlations than Avisoft-SASLab Pro. (G-I). For USV fundamental frequency, there were no significant within system differences and large correlations were found between assessments in each system, however, DeepSqueak again had higher overall correlations. The data points represent the mean and the error bars represent the standard error of the mean *** = p < .001.
When we assessed the within file consistency for duration in Avisoft-SASLab Pro, there was no significant difference between the first assessment and the second assessment (t (132) = .58, p = .57), nor was there a significant difference between the first and second assessment in DeepSqueak (t (132) = .44, p = .66) (Fig. 4D). In order to further clarify these relationships, correlations were run per system which found that there was a significant, large correlation between the first and second assessment of the same file in Avisoft-SASLab Pro (r (67) =.61, p < .0001), with an even larger correlation present between assessments for DeepSqueak (r (67) =.77, p < .0001) (Fig. 4E,F).
Lastly, we assessed the internal consistency for the fundamental frequency of USVs in each system. For Avisoft-SASLab Pro, we found no difference between the first and the second assessment (t (132) = .06, p = .96), nor was there a significant difference between assessments for DeepSqueak (t (132) = .10, p = .92) (Fig. 4G). When correlations were run to further parse out these relationships, we found that for Avisoft-SASLab Pro there was a significant, large correlation between the first and second assessment (r (67) =.83, p < .0001) (Fig. 4H). There was also a significant, even larger correlation between assessments for DeepSqueak (r (67) =.87, p < .0001) (Fig. 4I).
4. Discussion
The Avisoft microphone and recording interface is a gold standard in vocalization research, constituting a viable metric from which other recording programs can be measured. Spearheading Avisoft’s prominence is the degree of agency it confers the user, as it renders a two-dimensional representation of each vocalization in addition to transducing USVs to put them into the human auditory spectrum. Therefore, Avisoft-SASLab Pro allows an operator to accurately determine the validity of each detected vocalization via both visual and auditory cues, while granting the user the ability to remove any background noise that may be causing false positives. Thus, Avisoft-SASLab Pro is extremely adept at correctly detecting the total quantity of USVs emitted from an animal and, consequently, has been widely used for over 20 years. Recently, DeepSqueak, another easily accessible and accurate analysis system has risen to prominence. However, the two systems have not been directly compared, so their respective advantages and disadvantages are unknown.
Our study found that both Avisoft-SASLab Pro and DeepSqueak are highly accurate and consistent systems, as we observed correlations of .90, .97, and .99 across three different mouse strains when comparing the total quantity of USVs detected by the programs. Furthermore, no significant differences in the total USVs detected between DeepSqueak and Avisoft-SASLab Pro were found. The high reproducibility and consistency of results between Avisoft-SASLab Pro and DeepSqueak is significant, as findings are not always consistent between recording programs. For instance, in previous studies, we compared Avisoft-SASLab Pro to Ultravox 2.0 and Avisoft-SASLab Pro to the Mouse Song Analyzer and found that in both cases Avisoft-SASLab Pro was more accurate and correspondingly detected significantly more USVs than the other systems [3, 4]. We have also compared DeepSqueak to the Mouse Song Analyzer and similarly found that DeepSqueak was the more sensitive and thus the more accurate system [5]. Furthermore, Coffey et al. [6] compared DeepSqueak to both the Mouse Ultrasonic Profile ExTraction (MUPET) and Ultravox programs. They found that in both rats and mice DeepSqueak had the lowest miss rate and fewest false positives of any program evaluated [6]. Additionally, when DeepSqueak was assessed in environments with variable amounts of background noise, it was again more accurate than either Ultravox or MUPET [6]. Overall, the literature suggests that there can be significant discrepancies in USV detection between vocalization programs which may lead to problems with validity and reliability in communication research, however, our current study indicates that DeepSqueak avoids these pitfalls, as it comprises a consistent and effective tool for USV detection that is comparable to Avisoft-SASLab Pro.
We next compared the duration of time each program took to analyze a USV file. For files with low rates of USVs (200 or less), DeepSqueak took approximately 5 minutes to complete the file whereas Avisoft-SASLab Pro took approximately 15. For files that had high rates of USVs (over 200) DeepSqueak took 15 minutes to analyze the USVs whereas Avisoft-SASLab Pro took approximately 55. Therefore, in our hands, DeepSqueak was a faster USV analysis system than Avisoft-SASLab Pro. One factor that contributes to this disparity is that in Avisoft-SASLab Pro, call detection bars are generated around each call which are used to assess the call parameters, some of these are rendered accurately however others need to be trimmed in order to ensure accurate measurements. DeepSqueak has a similar limitation, in that automatic detection boxes are rendered around each call but not all are rendered accurately, requiring a user to redraw them to properly capture each aspect of the call. However, we found that more manual intervention was required in the Avisoft-SASLab Pro system than in DeepSqueak. Furthermore, the process of redrawing the detection box in DeepSqueak was a faster process than trimming the detection bars in Avisoft-SASLab Pro. Additionally, the Avisoft-SASLab Pro system requires the user to scroll through the entirety of the file. This is beneficial in that it eliminates the chance of false positives, guaranteeing accurate detection, however, it does slow down analysis times when compared to DeepSqueak which jumps from call to call. As an aside, it is likely that this feature of Avisoft-SASLab Pro makes it slightly more accurate than DeepSqueak, accounting for the non-significant differences in call detection between systems. Altogether, while both systems are highly accurate, DeepSqueak was a faster analysis program than Avisoft-SASLab Pro.
In addition to comparing Avisoft-SASLab Pro and DeepSqueak’s reliability for the total quantity of USVs detected, we also compared the average duration and fundamental frequency of the vocalizations across systems, other important aspects of communication that are commonly assessed [2, 10–12]. Across all strains, we observed predominately large correlations between Avisoft-SASLab Pro and DeepSqueak for the duration and fundamental frequency of the USVs. However, only one correlation met or exceeded the .9 standard for reliability, though several correlations were within .03 of the standard. Additionally, numerous significant differences were present between systems, particularly when assessing the average duration of the USVs. Therefore, when comparing Avisoft-SASLab Pro and DeepSqueak’s detection of the duration and fundamental frequency of USVs there are inconsistencies present, which may indicate that results attained from one program may not be attainable in the other.
We hypothesized that the variability between Avisoft-SASLab Pro and DeepSqueak for the spectral and temporal features of the USVs was most likely a product of subtle differences in the detection parameters between programs. As mentioned prior, we found that Avisoft-SASLab Pro required more manual adjustments to the automatically rendered detection boundaries than DeepSqueak. Since the process of adjusting call parameter bars is entirely mediated by the user, there may be slight discrepancies in the bounds set for each USV in a file. Therefore, more manual intervention introduces more possibility of variability, which could accumulate over time and lead to differences. In order to directly test this hypothesis, we assessed the internal validity of DeepSqueak and Avisoft-SASLab Pro by assessing then reassessing the same .wav file in each system. We found that while both systems had perfect call detection accuracy on repeat assessment, there were slight, non-significant differences when assessing the duration and fundamental frequency of the vocalizations in each system. Upon further analysis, DeepSqueak displayed higher correlations than Avisoft-SASLab Pro, indicating that DeepSqueak is more consistent when assessing spectral and temporal parameters, although the difference was minimal. Since each file may have several hundred USVs, even a slight discrepancy in USV parameter rendering may lead to lower than expected correlations both within and between systems, such as the one’s we observed in the current study. Furthermore, while it was rare, we did occasional encounter instances where the call parameter bars in Avisoft were unable to be reduced, skewing the measurements. This could also contribute to the differences in variability both within and between systems. In addition to the differences in duration and frequency between systems, explained by the above findings, we also observed a strain specific difference for duration, in that Avisoft-SASLab Pro detected calls of a shorter duration than DeepSqueak but only for C57 mice. Previously we have shown that there are no differences in background noise across FVB.129, FVB and C57 mice, however, we also found that C57 neonates emit softer vocalizations than their comparison strains [3]. Therefore, it is likely that the difference between systems for fundamental frequency in C57 pups is due to their respective softer vocalizations. It is possible that this feature is particularly sensitive to the loudness of a call, thereby accounting for the observed strain specific difference. While findings were not in 100% accordance between systems when assessing the spectral and temporal characteristics of USVs, it is important to reiterate that a repeat assessment of a file in both systems did not lead to any significant differences within the same file and large correlations were still observed between assessments, therefore, both systems do produce reliable data.
Due to the breadth of available USV analysis systems currently available, it is essential to ensure that each new system produces highly accurate and reliable results. In our study, we found that both Avisoft-SASLab Pro and DeepSqueak are accurate, accessible, and reliable systems, when assessing neonatal vocalizations in mice. Future studies could expand our work by examining adult vocalizations, as there are numerous differences between neonatal and adult USVs as well as the spectral and temporal dimensions of the calls [2, 13]. Comparisons between Avisoft-SASLab Pro and DeepSqueak could also be conducted in rat models to ensure that each system is equally accurate and reliable across species.
5. Conclusions
The present study compared the Avisoft-SASLab Pro analysis system to the DeepSqueak analysis system in order to assess the reproducibility of results between programs under congruent detection parameters. We assessed C57BL/6, FVB.129, and FVB mice since they comprise 52% of the most popular mouse strains purchased from Jackson labs and found large correlations between systems and no significant differences in the quantity of USVs that were detected. We also found predominately large correlations between systems when assessing the duration and fundamental frequency of USVs as well as several significant differences. Additionally, we determined that DeepSqueak analyzes USV files faster than Avisoft-SASLab Pro and is similarly accurate. Altogether, our study indicates that both DeepSqueak and Avisoft-SASLab Pro are accessible, reliable, and accurate USV analysis programs, constituting optimal tools for researching communicative behaviors.
Acknowledgments
The authors do not have any conflicts of interest to disclose. We would like to thank David Narvaiz, Paige Womble, and Greg Sullens for their critical review of the paper. Funding: This work was supported by the National Institutes of Health grant NS088776.
Footnotes
CRediT authorship contribution statement
Matthew S. Binder: Conceptualization, Writing – original draft, Writing – review & editing. Zachary J. Pranske: Conceptualization, Writing – original draft. Joaquin N. Lugo: Conceptualization, Writing – original draft, Writing – review & editing, Funding acquisition, Resources.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References
- [1].Esposito, & Venuti, Developmental changes in the fundamental frequency (f0) of infants’ cries: a study of children with Autism Spectrum Disorder, Early Child Dev. Care 180 (8) (2010) 1093–1102, 10.1080/03004430902775633. [DOI] [Google Scholar]
- [2].Scattoni ML, Gandhy SU, Ricceri L, Crawley JN, Unusual repertoire of vocalizations in the BTBR T+tf/J mouse model of autism, PLOS One 3 (8) (2008) e3067, 10.1371/journal.pone.0003067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Binder M, Nolan SO, Lugo JN, A comparison of the Avisoft (v.5.2) and MATLAB mouse song analyzer (v.1.3) vocalization analysis systems in C57BL/6, Fmr1-FVB.129, NS-Pten-FVB, and 129 mice, J. Neurosci. Methods 346 (2020), 108913, 10.1016/j.jneumeth.2020.108913. [DOI] [Google Scholar]
- [4].Binder MS, Hernandez-Zegada CJ, Potter CT, Nolan SO, Lugo JN, A comparison of the Avisoft (5.2) and Ultravox (2.0) recording systems: implications for early-life communication and vocalization research, J. Neurosci. Methods 309 (2018) 6–12, 10.1016/j.jneumeth.2018.08.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Binder MS, Pranske ZJ, Lugo JN, Evaluating the deepsqueak and mouse song analyzer vocalization analysis systems in C57BL/6J, FVB.129, and FVB neonates, J. f Neurosci. Methods 364 (2021), 109356, 10.1016/j.jneumeth.2021.109356. [DOI] [Google Scholar]
- [6].Coffey KR, Marx RG, Neumaier JF, DeepSqueak: a deep learning-based system for detection and analysis of ultrasonic vocalizations, Neuropsychopharmacology 44 (5) (2019) 859–868, 10.1038/s41386-018-0303-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Shekel I, Giladi S, Raykin E, Weiner M, Chalifa-Caspi V, Lederman D, Kofman O, Golan HM, Isolation-induced ultrasonic vocalization in environmental and genetic mice models of autism [Original Research], Front. Neurosci. 15 (2021), 10.3389/fnins.2021.769670. [DOI] [Google Scholar]
- [8].Wohr M, Roullet FI, Hung AY, Sheng M, Crawley JN, Communication ¨ impairments in mice lacking Shank1: reduced levels of ultrasonic vocalizations and scent marking behavior, PLOS One 6 (6) (2011) e20631, 10.1371/journal.pone.0020631. [DOI] [Google Scholar]
- [9].Shair HN, Acquisition and expression of a socially mediated separation response, Behav. Brain Res. 182 (2) (2007) 180–192, 10.1016/j.bbr.2007.02.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [10].Binder MS, Lugo JN, NS-Pten knockout mice show sex- and age-specific differences in ultrasonic vocalizations, Brain Behav. 7 (11) (2017) e00857, 10.1002/brb3.857. [DOI] [Google Scholar]
- [11].Esposito Hiroi, & Scattoni Cry, baby, cry: expression of distress as a biomarker and modulator in autism spectrum disorder, Int. J. Neuropsychopharmacol. 20 (6) (2017) 498–503, 10.1093/ijnp/pyx014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Scattoni ML, Crawley J, Ricceri L, Ultrasonic vocalizations: a tool for behavioural phenotyping of mouse models of neurodevelopmental disorders, Neurosci. Biobehav. Rev. 33 (4) (2009) 508–515, 10.1016/j.neubiorev.2008.08.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Scattoni ML, Ricceri L, Crawley JN, Unusual repertoire of vocalizations in adult BTBR T+tf/J mice during three types of social encounters, Genes, Brain Behav. 10 (1) (2011) 44–56, 10.1111/j.1601-183X.2010.00623.x. [DOI] [PMC free article] [PubMed] [Google Scholar]