Shashika Udara’s Post

Tuning Word2Vec with Bayesian Optimization: Applied to Music Recommendations via #TowardsAI → https://bit.ly/3Q5mY8T

Tuning Word2Vec with Bayesian Optimization: Applied to Music Recommendations via #TowardsAI →

I’ve been working on an LSTM based RNN model that composes music. This tool takes a creative approach to generating melodies in abc notation, converting them into playable audio, shifting the task from music generation to NLP and making it accessible through an easy-to-use interface. Key Features: 🎼 Generates original music in ABC notation 🎧 Converts the notation into high-quality audio 🖥️ Offers an interactive Gradio interface for seamless usage Would love to hear your thoughts or feedback! If you’re as passionate about Neural Networks and music as I am, let’s connect and discuss new ideas. #AI #Music #Innovation #DeepLearning #AIMusic https://lnkd.in/gyp8cgKu https://lnkd.in/gwDae_-F

Music genre classification with deep learning involves using algorithms to automatically categorize music into different genres based on its audio features. Overfitting is a common issue where the model performs well on the training data but poorly on new data. To address this, we use techniques like dropout and regularization to help the model generalize better. We train the model on a dataset of labeled audio files, validate it on a separate set, and test it on unseen data. By understanding these principles, we can build models that accurately classify music genres, opening up new possibilities for music analysis and recommendation systems.

Generate high-quality music from text with Open Music using standard GPU memory and just a few lines of code. Learn how with a simple, step-by-step tutorial. #AI #LLM #opensource

Celebrating My First Machine Learning Project Milestone! 🎉 I'm overjoyed to share that I've successfully completed my first project in the exciting field of machine learning! This project, focused on music genre classification using various techniques, has been an incredible learning experience. Throughout this journey, I delved deep into feature extraction methods, exploring both time and frequency domains to capture the essence of audio signals. Techniques like MFCC, Chromagram, and Spectral Centroid enabled me to uncover valuable insights from the data. The unsupervised learning phase involved employing the K-Means algorithm for clustering and Principal Component Analysis (PCA) for dimensionality reduction and visualization. Visualizing the feature space and interpreting the results was truly fascinating! In the supervised learning phase, I trained and evaluated multiple classifiers, including K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest, and Neural Networks. It was remarkable to witness their different strengths and weaknesses in tackling this multi-class classification problem. While the road wasn't always smooth, I learned invaluable lessons about model evaluation, interpretation of results, and the importance of techniques like data augmentation for improving accuracy. I'm thrilled to have explored various evaluation metrics, such as confusion matrices, ROC curves, and area under the curve (AUC), which provided deeper insights into the performance of my models. This project has not only equipped me with practical skills but has also ignited my passion for machine learning even further. I'm excited to continue learning, growing, and tackling more complex challenges in this dynamic field. I would like to express my gratitude to the incredible machine learning community for their invaluable resources and support, which have been instrumental in my learning journey. Here's to many more milestones and achievements in the world of machine learning! 🚀 If you want to have a look check my repo on GitHub: