Jorge Perianez-Pascual

This paper presents a novel approach to audio classification leveraging the latent representation generated by Meta’s EnCodec neural audio codec. We hypothesize that the compressed latent space representation captures essential audio features more suitable for classification tasks than the traditional spectrogram-based approaches. We train a vanilla convolutional neural network for music genre, speech/music, and environmental sound classification using EnCodec’s encoder output as input to validate this. Then, we compare its performance training with the same network using a spectrogram-based representation as input. Our experiments demonstrate that this approach achieves comparable accuracy to state-of-the-art methods while exhibiting significantly faster convergence and reduced computational load during training. These findings suggest the potential of EnCodec’s latent representation for efficient, faster, and less expensive audio classification applications. We analyze the characteristics of EnCodec’s output and compare its performance against traditional spectrogram-based approaches, providing insights into this novel approach’s advantages.

The popularity of bots is on the rise, with many bots able to interact with users via a chat or voice interface thanks to the embedding of a Natural Language Processing (NLP) component. Still, companies often express concerns about the quality of such bots, as their malfunctioning could have a severe impact on the company revenue or image. Unfortunately, the field of testing NLP-intensive bots is still in its infancy. This paper aims to characterize the testing properties and techniques (and their adaptation) relevant to this type of bots. We believe this will be helpful as a reference framework to compare and evaluate future bot testing research initiatives.

OCR engines aim to identify and extract text strings fromdocuments or images. While current efforts focus mostly inmainstream languages, there is little support for program-ming or domain-specific languages (DSLs). In this paper, wepresent our vision about the current state of OCR recognitionfor DSLs and its challenges. We discuss some strategies toimprove the OCR quality applied to DSL textual expressionsby leveraging DSL specifications and domain data. To bettersupport our ideas we present the preliminary results of anempirical study and outline a research roadmap.