Mobile app for targeted selective treatment of haemonchosis in sheep.

Livestock is an important part of many countries gross domestic product, and sanitary control impacts herd management costs. To contribute to incorporating new technologies into this economic chain, this work presents a mobile application for decision assistance to treatment against parasitic infection by Haemonchus contortus in small ruminants. Based on the Android system, the proposed software is a semi-automated computer-aided procedure to assist Famacha© pre-trained farmers in applying anthelmintic treatment. It mimics the two-class decision procedure performed by the veterinarian with the help of the Famacha© card. The embedded cell phone camera was employed to acquire an image from the ocular conjunctival mucosa, classifying the animal as healthy or anemic. Two machine-learning strategies were assessed, resulting in an accuracy of 83 % for a neural network and 87 % for a support vector machine (SVM). The SVM classifier was embedded into the app and made available for evaluation. This work is particularly interesting to small property owners from regions with difficult access or restrictions on obtaining continuous post-training technical guidance to use the Famacha© method effectively.

Online approximation of the multichannel Wiener filter with preservation of interaural level difference for binaural hearing-aids.

This work presents an online approximation method for the multichannel Wiener filter (MWF) noise reduction technique with preservation of the noise interaural level difference (ILD) for binaural hearing-aids. The steepest descent method is applied to a previously proposed MWF-ILD cost function to both approximate the optimal linear estimator of the desired speech and keep the subjective perception of the original acoustic scenario. The computational cost of the resulting algorithm is estimated in terms of multiply and accumulate operations, whose number can be controlled by setting the number of iterations at each time frame. Simulation results for the particular case of one speech and one-directional noise source show that the proposed method increases the signal-to-noise ratio SNR of the originally acquired speech by up to 16.9 dB in the assessed scenarios. As compared to the online implementation of the conventional MWF technique, the proposed technique provides a reduction of up to 7 dB in the noise ILD error at the price of a reduction of up 3 dB in the output SNR. Subjective experiments with volunteers complement these objective measures with psychoacoustic results, which corroborate the expected spatial preservation of the original acoustic scenario. The proposed method allows practical online implementation of the MWF-ILD noise reduction technique under constrained computational resources. Predicted SNR improvements from 12 dB to 16.9 dB can be obtained in application-specific integrated circuits for hearing-aids and state-of-the-art digital signal processors.

Theoretical transient analysis of a hearing aid feedback canceller with a saturation type nonlinearity in the direct path.

Feedback cancellation in a hearing aid is essential for achieving high maximum stable gain to compensate for the losses in severe to profound hearing impaired people. The performance of adaptive feedback cancellers has been studied by assuming that the feedback path can be modeled as a linear system. However, limited dynamic range, low-cost loudspeakers, and nonlinear power amplifiers may distort the hearing aid output signal. In this way, linear-based predictions of the canceller performance may lead to significant deviations from its actual behavior. This work presents a theoretical performance analysis of a Least Mean Square based shadow filter that is applied to set up the coefficients of a feedback canceller, which is subject to a static saturation type nonlinearity at the output of the direct path. Deterministic recursive equations are derived to predict the mean square feedback error and the mean coefficient vector evolution between updates of the feedback canceller. These models are defined as functions of the canceller parameters and input signal statistics. Comparisons with Monte Carlo simulations show the provided models are highly accurate under the considered assumptions. The developed models allow inferences about the potential impact of an overdriven loudspeaker over the transient performance of the direct method feedback canceller, serving as insightful tools for understanding the involved mechanisms.

A feed forward adaptive canceller to reduce the occlusion effect in hearing aids.

Hearing aids are essential devices for social integration of hearing impaired people in order to improve their auditory perception. Recent studies have reported significant dissatisfaction factors that tend to reduce their daily use. The occlusion effect is one important source of complaints. This phenomenon stems from the partial or complete closure of the ventilation opening of the ear-mould, usually performed to prevent feedback effects in high-gain devices. This work presents a new adaptive active-noise-control system to reduce the occlusion effect in small- or unvented hearing aids. In contrast to previously developed occlusion-effect cancellers, this system offers a feedforward cancelling structure that permits the analysis of its behaviour as a finite-impulse-response linear-filter identification problem. Deterministic recursive equations were derived with the aim to theoretically predict its mean square error and mean coefficient behaviour, both in transient and steady state conditions. Such models are of particular interest to hearing aid designers as guide tools for setting parameters to obtain a desired performance. Computational simulations accurately agree with theoretical predictions obtained by the derived equations, indicating a mean reduction of 5.4dB of the occlusion effect in the range of 200-500Hz. Subjective experiments with the use of a real prototype corroborate the functionality of the proposed architecture. No perceptual side effects regarding high-frequency amplifications of the original sounds were reported by volunteers.