Humans have been entertained by music for millennia. For ages it has been treated as an art form which requires a lot of imagination, creativity and accumulation of feelings and emotions. Recent trends in the field of Artificial Intelligence have been getting traction and Researchers have been developing and generating rudimentary forms of music through the use of AI. Our goal is to generate novel music, which will be non-repetitive and enjoyable. We aim to utilize a couple of Machine Learning models for the same. Given a seed bar of music, our first Discriminatory network consisting of Support Vector Machines and Neural Nets will choose a note/chord to direct the next bar. Based on this chord or note another network, a Generative Net consisting of Generative Pretrained Transformers(GPT2) and LSTMs will generate the entire bar of music. Our two fold method is novel and our aim is to make the generation method as similar to music composition in reality as possible. This in turn results in better concordant music. Machine generated music will be copyright free and can be generated conditioned on a few parameters for a given use.The paper presents several use cases and while the utilization will be for a niche audience, if a easy to use application can be built, almost anyone will be able to use deep learning to generate concordant music based on their needs.
Sign language detection, identifying if someone is signing or not, is becoming crucially important for its applications in remote conferencing software and for selecting useful sign data for training sign language recognition or translation tasks. We argue that the current benchmark data sets for sign language detection estimate overly positive results that do not generalize well due to signer overlap between train and test partitions. We quantify this with a detailed analysis of the effect of signer overlap on current sign detection benchmark data sets. Comparing accuracy with and without overlap on the DGS corpus and Signing in the Wild, we observed a relative decrease in accuracy of 4.17% and 6.27%, respectively. Furthermore, we propose new data set partitions that are free of overlap and allow for more realistic performance assessment. We hope this work will contribute to improving the accuracy and generalization of sign language detection systems.
Recent approaches in Title generation using neural approaches have relied on an end to end deep learning system based on the sequence to sequence model. Such approaches have yielded good results but remain constricted in use due to a fixed size input which is often very small compared to the text being used or might take huge compute power to train and use if input size is increased. Our approach amalgamates an extractive and abstractive approach to get the best of both worlds using a textrank algorithm for the extractive part and a reasonably small seq2seq architecture as the abstractive part. Testing on the Amazon Fine Food Review dataset, our approach gives good results using less compute power.We utilize the prevailing metrics of ROUGE and Cosine Similarity. Manual checking shows that the majority of our generated topics are grammatically correct.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.