A novel deep transfer learning models for recognition of birds sounds in different environment

Automatic species classification of birds from their sound is a computational tool of increasing importance in ecology, conservation monitoring and vocal communication studies. To make classification useful in practice, it is crucial to improve its accuracy while ensuring that it can run at big data scales. Many approaches use acoustic measures based on spectrogram-type data, such as the Mel-frequency cepstral coefficient (MFCC) features which represent a manually-designed summary of spectral information. However, recent work in machine learning has demonstrated that features learnt automatically from data can often outperform manually-designed feature transforms. Feature learning can be performed at large scale and “unsupervised”, meaning it requires no manual data labelling, yet it can improve performance on “supervised” tasks such as classification. In this work we introduce a technique for feature learning from large volumes of bird sound recordings, inspired by techniques that have proven useful in other domains. We experimentally compare twelve different feature representations derived from the Mel spectrum (of which six use this technique), using four large and diverse databases of bird vocalisations, classified using a random forest classifier. We demonstrate that in our classification tasks, MFCCs can often lead to worse performance than the raw Mel spectral data from which they are derived. Conversely, we demonstrate that unsupervised feature learning provides a substantial boost over MFCCs and Mel spectra without adding computational complexity after the model has been trained. The boost is particularly notable for single-label classification tasks at large scale. The spectro-temporal activations learned through our procedure resemble spectro-temporal receptive fields calculated from avian primary auditory forebrain. However, for one of our datasets, which contains substantial audio data but few annotations, increased performance is not discernible. We study the interaction between dataset characteristics and choice of feature representation through further empirical analysis.

There has been great expectation that the knowledge of an individual's genotype will provide a basis for assessing susceptibility to diseases and designing individualized therapy. Non-synonymous single nucleotide polymorphisms (nsSNPs) that lead to an amino acid change in the protein product are of particular interest because they account for nearly half of the known genetic variations related to human inherited diseases. To facilitate the identification of disease-associated nsSNPs from a large number of neutral nsSNPs, it is important to develop computational tools to predict the phenotypic effects of nsSNPs. We prepared a training set based on the variant phenotypic annotation of the Swiss-Prot database and focused our analysis on nsSNPs having homologous 3D structures. Structural environment parameters derived from the 3D homologous structure as well as evolutionary information derived from the multiple sequence alignment were used as predictors. Two machine learning methods, support vector machine and random forest, were trained and evaluated. We compared the performance of our method with that of the SIFT algorithm, which is one of the best predictive methods to date. An unbiased evaluation study shows that for nsSNPs with sufficient evolutionary information (with not <10 homologous sequences), the performance of our method is comparable with the SIFT algorithm, while for nsSNPs with insufficient evolutionary information (<10 homologous sequences), our method outperforms the SIFT algorithm significantly. These findings indicate that incorporating structural information is critical to achieving good prediction accuracy when sufficient evolutionary information is not available. The codes and curated dataset are available at http://compbio.utmem.edu/snp/dataset/