DCNN for Pig Vocalization and Non-Vocalization Classification: Evaluate Model Robustness with New Data

Since pig vocalization is an important indicator of monitoring pig conditions, pig vocalization detection and recognition using deep learning play a crucial role in the management and welfare of modern pig livestock farming. However, collecting pig sound data for deep learning model training takes time and effort. Acknowledging the challenges of collecting pig sound data for model training, this study introduces a deep convolutional neural network (DCNN) architecture for pig vocalization and non-vocalization classification with a real pig farm dataset. Various audio feature extraction methods were evaluated individually to compare the performance differences, including Mel-frequency cepstral coefficients (MFCC), Mel-spectrogram, Chroma, and Tonnetz. This study proposes a novel feature extraction method called Mixed-MMCT to improve the classification accuracy by integrating MFCC, Mel-spectrogram, Chroma, and Tonnetz features. These feature extraction methods were applied to extract relevant features from the pig sound dataset for input into a deep learning network. For the experiment, three datasets were collected from three actual pig farms: Nias, Gimje, and Jeongeup. Each dataset consists of 4000 WAV files (2000 pig vocalization and 2000 pig non-vocalization) with a duration of three seconds. Various audio data augmentation techniques are utilized in the training set to improve the model performance and generalization, including pitch-shifting, time-shifting, time-stretching, and background-noising. In this study, the performance of the predictive deep learning model was assessed using the k-fold cross-validation (k = 5) technique on each dataset. By conducting rigorous experiments, Mixed-MMCT showed superior accuracy on Nias, Gimje, and Jeongeup, with rates of 99.50%, 99.56%, and 99.67%, respectively. Robustness experiments were performed to prove the effectiveness of the model by using two farm datasets as a training set and a farm as a testing set. The average performance of the Mixed-MMCT in terms of accuracy, precision, recall, and F1-score reached rates of 95.67%, 96.25%, 95.68%, and 95.96%, respectively. All results demonstrate that the proposed Mixed-MMCT feature extraction method outperforms other methods regarding pig vocalization and non-vocalization classification in real pig livestock farming.

Chronic recording of cortical activity underlying vocalization in awake minipigs

BACKGROUND

Investigating brain dynamics underlying vocal production in animals is a powerful way to inform on the neural bases of human speech. In particular, brain networks underlying vocal production in non-human primates show striking similarities with the human speech production network. However, despite increasing findings also in birds and more recently in rodents, the extent to which the primate vocal cortical network model generalizes to other non-primate mammals remains unclear. Especially, no domestic species has yet been proposed to investigate vocal brain activity using electrophysiological approaches.

NEW METHOD

In the present study, we introduce a novel experimental paradigm to identify the cortical dynamics underlying vocal production in behaving minipigs. A key problem to chronically implant cortical probes in pigs is the presence and growth of frontal sinuses extending caudally to the parietal bone and preventing safe access to neural structures with conventional craniotomy in adult animals.

RESULTS

Here we first show that implantations of soft ECoG grids can be done safely using conventional craniotomy in minipigs younger than 5 months, a period when sinuses are not yet well developed. Using wireless recordings in behaving animals, we further show activation of the motor and premotor cortex around the onset of vocal production of grunts, the most common vocalization of pigs.

CONCLUSION

These results suggest that minipigs, which are very loquacious and social animals, can be a good experimental large animal model to study the cortical bases of vocal production.

Ontogeny of individual and litter identity signaling in grunts of piglets

Many studies have shown that animal vocalizations can signal individual identity and group/family membership. However, much less is known about the ontogeny of identity information-when and how this individual/group distinctiveness in vocalizations arises and how it changes during the animal's life. Recent findings suggest that even species that were thought to have limited vocal plasticity could adjust their calls to sound more similar to each other within a group. It has already been shown that sows can acoustically distinguish their own offspring from alien piglets and that litters differ in their calls. Surprisingly, individual identity in piglet calls has not been reported yet. In this paper, this gap is filled, and it is shown that there is information about piglet identity. Information about litter identity is confirmed as well. Individual identity increased with age, but litter vocal identity did not increase with age. The results were robust as a similar pattern was apparent in two situations differing in arousal: isolation and back-test. This paper argues that, in piglets, increased individual discrimination results from the rapid growth of piglets, which is likely to be associated with growth and diversification of the vocal tract rather than from social effects and vocal plasticity.

Encoding of situations in the vocal repertoire of piglets (Sus scrofa): a comparison of discrete and graded classifications

Two important questions in bioacoustics are whether vocal repertoires of animals are graded or discrete and how the vocal expressions are linked to the context of emission. Here we address these questions in an ungulate species. The vocal repertoire of young domestic pigs, Sus scrofa, was quantitatively described based on 1513 calls recorded in 11 situations. We described the acoustic quality of calls with 8 acoustic parameters. Based on these parameters, the k-means clustering method showed a possibility to distinguish either two or five clusters although the call types are rather blurred than strictly discrete. The division of the vocal repertoire of piglets into two call types has previously been used in many experimental studies into pig acoustic communication and the five call types correspond well to previously published partial repertoires in specific situations. Clear links exist between the type of situation, its putative valence, and the vocal expression in that situation. These links can be described adequately both with a set of quantitative acoustic variables and through categorisation into call types. The information about the situation of emission of the calls is encoded through five call types almost as accurately as through the full quantitative description.

Importance of mother-infant communication for social bond formation in mammals

Mother-infant bonding is a universal relationship of all mammalian species. Here, we describe the role of reciprocal communication between mother and infant in the formation of bonding for several mammalian species. Mother-infant bond formation is reinforced by various social cues or stimuli, including communicative signals, such as odor and vocalizations, or tactile stimuli. The mother also develops cross-modal sensory recognition of the infant, during bond formation. Many studies have indicated that the oxytocin neural system plays a pivotal role in bond formation by the mother; however, the underlying neural mechanisms for infants have not yet been clarified. The comparative understanding of cognitive functions of mother and infants may help us understand the biological significance of mother-infant communication in mammalian species.

Vocal and locomotor responses of piglets to social isolation and reunion

Potentiation of infant isolation calls following a brief reunion with the mother is considered an index of filial bonding in altricial rodents. We investigated potentiation of isolation and reunion responses in 15-day-old unweaned domestic piglets (Sus scrofa domesticus). When piglets were re-isolated following a brief, comforting reunion with their mother and littermates in their home pen, they displayed a persistence (relative potentiation) of calling and jumping. In contrast, when re-isolated following a brief interaction with their mother or an unfamiliar sow in a familiar or unfamiliar location, or with littermates alone, calling rate and locomotion dropped. Subsequently, piglets spent more time near their mother's face if they had previously interacted with an unfamiliar sow rather than their mother, consistent with maternal recognition. Although we did not detect maternal potentiation as described in altricial rodents, filial attachment bonding was clearly evident in piglet responses, especially during reunions.

Asymmetries in the individual distinctiveness and maternal recognition of infant contact calls and distress screams in baboons

A key component of nonhuman primate vocal communication is the production and recognition of clear cues to social identity that function in the management of these species' individualistic social relationships. However, it remains unclear how ubiquitous such identity cues are across call types and age-sex classes and what the underlying vocal production mechanisms responsible might be. This study focused on two structurally distinct call types produced by infant baboons in contexts that place a similar functional premium on communicating clear cues to caller identity: (1) contact calls produced when physically separated from, and attempting to relocate, mothers and (2) distress screams produced when aggressively attacked by other group members. Acoustic analyses and field experiments were conducted to examine individual differentiation in single vocalizations of each type and to test mothers' ability to recognize infant calls. Both call types showed statistically significant individual differentiation, but the magnitude of the differentiation was substantially higher in contact calls. Mothers readily discriminated own-offspring contact calls from those of familiar but unrelated infants, but did not do so when it came to distress screams. Several possible explanations for these asymmetries in call differentiation and recognition are considered.