Download Objective Evaluations of Synthesised Environmental Sounds There are a range of different methods for comparing or measuring the similarity between environmental sound effects. These methods can be used as objective evaluation techniques, to evaluate the effectiveness of a sound synthesis method by assessing the similarity between synthesised sounds and recorded samples. We propose to evaluate a number of different synthesis objective evaluation metrics, by using the different distance metrics as fitness functions within a resynthesis algorithm. A recorded sample is used as a target sound, and the resynthesis is intended to produce a set of synthesis parameters that will synthesise a sound as close to the recorded sample as possible, within the restrictions of the synthesis model. The recorded samples are excerpts of selections from a sound effects library, and the results are evaluated through a subjective listening test. Results show that one of the objective function performs significantly worse than several others. Only one method had a significant and strong correlation between the user perceptual distance and the objective distance. A recommendation of an objective evaluation function for measuring similarity between synthesised environmental sounds is made.
Download End-to-end equalization with convolutional neural networks This work aims to implement a novel deep learning architecture to perform audio processing in the context of matched equalization. Most existing methods for automatic and matched equalization show effective performance and their goal is to find a respective transfer function given a frequency response. Nevertheless, these procedures require a prior knowledge of the type of filters to be modeled. In addition, fixed filter bank architectures are required in automatic mixing contexts. Based on end-to-end convolutional neural networks, we introduce a general purpose architecture for equalization matching. Thus, by using an end-toend learning approach, the model approximates the equalization target as a content-based transformation without directly finding the transfer function. The network learns how to process the audio directly in order to match the equalized target audio. We train the network through unsupervised and supervised learning procedures. We analyze what the model is actually learning and how the given task is accomplished. We show the model performing matched equalization for shelving, peaking, lowpass and highpass IIR and FIR equalizers.