Cattle breeds in India consist of a mix of indigenous and exotic breeds. Jersey and Holstein Friesian are exotic breeds, whereas Sindhi, Sahiwal, and Gir are significant indigenous breeds. Researchers have observed that the cattle vocalisation is a good expression of their emotional well-being. We conducted two groups of experiments to collect cattle behaviour datasets and hunger intensity datasets. The first one was to map cattle vocalisation from some commonly found breeds in India to potential cattle intents or behaviours. The second dataset was from the delayed feed experiment to understand the relationship between the cattle vocalisation intensity levels and the potential hunger behaviour. The cattle behaviour dataset had 120 utterances, corresponding to six behaviours while hunger intensity dataset had 302 utterances for various hunger levels. These behaviours and intensity levels were labelled by the domain experts familiar with the cattle behaviour. The dataset was scaled and augmented with four different methods to generate 870 cattle sounds for the six classes and 1510 cattle sounds for three hunger intensity levels. We used MFCCs and OpenSMILE global features from the audio signal with 6552 properties to develop models. We tested two model architectures for behaviour classification. These models were tested on unseen cattle sounds for speaker-independent verification. Further, we compared different models with OpenSMILE features to detect hunger intensity levels. We achieved \(97\%\) accuracy for behaviour classification and \(89\%\) for hunger intensity level classification, validating that cattle vocalisation from Indian breeds can be used for cattle behaviour recognition with high confidence.