Abstract: Liquid-crystal (LC) optically addressable spatial light modulators (OASLMs) allow control of the phase and amplitude of optical beams. By performing wave mixing in an OASLM, we show that coherent phase detection can be achieved for light beams passing through highly scattering media, such as foam layers with several cm thicknesses. Thanks to the adaptive response of our OASLM, the phase information on the speckle signal is transferred at the output of the OASLM to the plane wave reference beam, allowing the cleaning of optical distortions and the direct measurement of amplitude phase modulations with a small diameter single photodiode. A good signal-to-noise ratio (SNR) is demonstrated for foam thickness up to 3 cm. These properties, together with the recently demonstrated sub-ms response time of our OASLM, make the method compatible with foreseen applications for imaging in biomedical tissues and turbid media.

Abstract: Bioacoustics is one of the most popular methods in bat research. Bat species are identifiable through their echolocation call features (e.g. peak frequency, duration, bandwidth) but the amounts of recordings to process generally require the help of machine learning algorithms. Yet, classifiers are only developed in some areas of the world and it may take dozens of years before they are available everywhere because reference calls are still lacking for numerous species. Our goal was to develop a universal classifier that would classify bat sonotypes according to call shape and peak frequency. To achieve this, we first defined eight sonotype categories that cover all bat echolocation shapes worldwide. We then trained a classifier using random forest decision trees with a database of 1,154,835 labelled sound events containing bat and non-bat sounds from four continents. After classification, we developed a process to group detected sound events according to the probability scores of their predicted sonotype category and their peak frequency. We then tested the performance of our classifier on a different set of recordings originating from five continents. Depending on the bat sonotype tested, the performance (area under ROC curve) of our classifier ranged between 0.77 and 0.99 for low-quality calls (SNR < 25 dB). Performance ranged between 0.89 and 1 for middle- or high-quality calls (SNR ≥ 25 dB). The performance for bat feeding buzz classification ranged between 0.93 and 0.98 depending on the SNR. The classifier was not developed to classify bat social calls; the majority of them were classified as a bat sonotype. The classifier is an open data format and can be used by anyone to study bats around the world. It can be used to spot acoustically described species but for which a classifier was not developed, and even to detect species that were not acoustically described yet. The grouping of sound events according to call sonotype and peak frequency may be used to describe bat communities and compare the composition of acoustic niches across time and space. This allows the monitoring of bats and the assessment of bat conservation issues in any region of the world.